Subcutaneous Transmitter Development

© 2013-2017, Kevan Hashemi, Open Source Instruments Inc.

Contents

2013
SEP-13OCT-13NOV-13DEC-13
2014
JAN-14FEB-14MAR-14APR-14
MAY-14JUN-14JUL-14AUG-14
SEP-14OCT-14NOV-14DEC-14
2015
JAN-15FEB-15MAR-15APR-15
MAY-15JUN-15JUL-15AUG-15
SEP-15OCT-15NOV-15DEC-15

2013

SEP-13

[19-SEP-13] Receive 10 assembled A3028 circuit boards. Measure quiescent current with external battery and power switched off to circuit. We program all ten with their channel numbers, channel X enabled at 512 SPS, channel Y disabled. We calibrate the ring oscillators and find that the bit rate is 200±5 ns for tcd_divisor = 24 for all ten. We set frequency_low to 8 for all transmitters. We cannot calibrate the center frequencies today because our spectrometer is broken. We obtain over 90% reception with the MAX2624 held against a receiving antenna. All transmitters detect mains hum. Current consumption results in table below.


Figure: Current Consumption of A3028As.

We solder a battery to the board. We find that P3-4 is not connected to 0V, so we connect it with a wire. With a jumper across P3-3 and P3-4, we can switch on and off the circuit. But without this jumper, R1 keeps transistor U1-1 turned off so the battery is disconnected from the circuit. The only current flowing is through 10 MΩ to 0V, which is around 0.27 μA.

[20-SEP-13] Create the first two-channel 512 SPS transmitter: the calibrated A3028A with leads shown here. Operating current is 132 μA. We re-program to create a one-channel 512 SPS transmitter and operating current is 76 μA. Thus we have a 20-μA base current plus 0.11 μA/SPS. We test both inputs of the A3028A and find they track almost exactly one another when we apply the same mains hum or heart beat. Input noise is 20 counts rms on each, which is only 8 μV rms.

Disabling the three test point outputs drops the operating current by 2±1 μA for 512 SPS and 1024 SPS. We resolve not to disable the test points because doing so complicates the calibration process.

[24-SEP-13] We load a battery into prototype A3028A transmitter 32.2, which has channel numbers 2 and 3. We apply acrylic conformal coating. We must jumper P3-3 to 0V to get the battery to connect. We measure the frequency response of both channels, as shown here. We clip off the programming extension. The transmitter now turns on and off, and picks up mains hum. We encapsulate in black epoxy and leave to cure.

OCT-13

[25-OCT-13] Four out of our nine un-encapsulated A3028As will not switch on. We trace the problem to the BGA-5 package of U3. We re-heat these chips, but they simply come off on the iron. Their solder blobs appear to have broken off the package. Since we first tested the circuits we have applied acrylic coating and loaded batteries. Several weeks have gone by. In a loupe, we see the ball of U3-2 appears to be convex at the package, as if it has broken off. Torquing the board during depanelization might be the cause of such a problem, or a flaw in the package construction.


Figure: BGA-5 Solder Ball Convex at Top. These particular chips are working. We already repaired the broken ones. But they look similar to the broken chips.

We send the above photographs to the company that assembled the circuit to see if they have any ideas about what went wrong. We replaced U3 on 32.4, 32.6, 33.2, and 33.4. The others, 32.8, 32.10, and 32.12, which were not faulty, we leave as they are. We program all of them as A3028As. Current consumption is 2.5±0.2 μA when inactive and 145±5 μA when awake.

[28-OCT-13] Our No32.2 encapsulated A3028A has been in water for over three weeks. The gain of both channels is nominal through the pass-band, and in agreement to within ±0.5 dB or better.

[31-OCT-13] At ION, we find that No32.2 transmits only 256 SPS per channel. When we measured its frequency response, we were measuring the response of its low-pass filters, but above 128 Hz we must have been seeing an aliased version of the input sinusoid.

NOV-13

[05-NOV-13] We have seven encapsulated A3028As and A3028Ds. Their programming extensions are still attached. If we apply the jumper between P3-3 and P3-4, we can run the transmitter of its own battery. We remove the jumper and apply an external battery through an ammeter and measure the sleeping and active current consumption. If we run one of the A3028As for several minutes minutes using its own battery, and then switch to using an external battery, the sleep current immediately after the switch is 80 μA. Within thirty seconds the sleep current has dropped to 40 μA. Within ten minutes, it is back to the original 2.1 μA. We do not observe this jump in sleep current with the A3028D.

The A3028A is equipped with the BR1225 battery. When we draw 145 μA from this battery, its voltage drops from 2.8 V to around 2.6 V. Our external battery is 3.0 V. We suspect that the U1 p-channel mosfet's parasitic diode is conducting tens of microamps from the external to the internal battery while the internal battery recovers to 2.8 V.

[18-NOV-13] We have A3028A transmitters 33.2, 33.4, and 33.6 in water. They measure 14.0 mm x 13.5 mm x 7.4 mm, with variations due to lumps, but these are the average values as best as we can guess. This puts the volume at less than 1.4 ml.

[20-NOV-13] We have A3028D transmitters 32.6, 32.8, 32.10, and 32.12 encapsulated. We measure their frequency response, and find it most satisfactory. We put them in water.

[25-NOV-13] Transmitters 32.6, 32.8, 32.10, 32.12, 33.2, 33.4, and 33.6 have been soaking in water for at least four days. All these were made with acrylic coating on the amplifiers while masking the battery pads, then soldering the battery with no-clean flux. We turn them on, put them in our faraday enclosure while still in water and record the input noise. A3028Ds have average input values around 45k, while A3028As have average around 47k. This suggests battery voltage 2.6 V for the BR2330 batteries on the A3028Ds and 2.5 V for the BR1225 batteries on the A3028As. The temperature in our office is around 17°C, so these values are consistent with the battery data sheets.

We remove from water and lay on a towel in the faraday enclosure. We see no square waves. We measure frequency response. The gain of each channel is always within 0.2 dB of its partner in the same transmitter. Soon after turning on our function generator and removing the transmitters form water, we measure the gain of all transmitters to be 2 dB too hight from 1-20 Hz. Half an hour later, all transmitters have the same gain we measured before the soak. We cool the transmitter with freezer spray and heat them in hot water, but observe no change in their frequency response.

DEC-13

[18-DEC-13] A few weeks ago we received 100 of A3028 assembled circuits from an assembly company. Out of the bag, one in five will not stay switched on. This is the same problem we observed on 25-OCT-13. We take 17 circuits that work and put them through four cycles of 5°C-60°C. One of the seventeen does not work at the end. The faulty circuits will work if we push down on U3, the BGA-5, with a stick. We replace U3 on two faulty circuits and they both work afterwards. We send six faulty circuits back to the assembly company, along with apparatus for observing the fault with the switch. We calibrate sixteen working circuits and load wires, antennas, and batteries. Today thirteen out of sixteen have the switch problem. We replace U3 on five of them, and they all work now. We photographed one of the BGA-5s we removed, next to its footprint.


Figure: BGA-5 Removed and Inverted Next to Footprint.

In the photograph we see two balls with flat tops on the footprint. On the component there are three pads with solder residue and two with no solder residue. These two match the flat-topped solder balls. We suppose that the balls broke away from the package during electronic assembly and subsequently failure takes place when oxide builds up between the two surfaces. We resolve to replace all the BGA-5s before calibration.

From ION, we receive the first report from Rob Wykes of recordings made with an A3028D implanted in a rat, see here.

[20-DEC-13] We replace U3 on another eleven A3028 circuits. All fo them now work perfectly. In three transmitters we found problems arising from excessive solder on the battery joints.

[24-DEC-13] Transmitters 32.2 and 32.6 have been in water for over five weeks and show no sign of the square wave problem. These were made with acrylic coating and no-clean flux for the battery terminals.

[31-DEC-13] We modify the P3028A01.abl firmware so that the transmit clock (TCK) waveform mark-space ratio is always 50%. We create P3028A02.abl. We do this at the expense of resolution in the period. With transmitter 34.7 we measure TCK period with divisor.


Figure: Transmit Clock Period versus Fast Clock Divisor. Firmware V2. The nominal period is 200 ns for 5 MHz bit rate.

We cool a transmitter with freezer spray. Its internal temperature drops to around −20°C. Its transmit clock period drops from 207 ns to 200 ns. This suggests of order 1% drop in the logic propagation delay per 10°C drop in temperature.

We test 4 A3028Ds and 11 A3028As with epoxy and varnish encapsulation. All have their programming extensions. We power them with their own batteries by means of a jumper from !ON to 0V. We attempt turn them on and off with a magnet half a dozen times each. One transmitter, A3028D No36.1, will not turn off. All others turn on and off normally. In two transmitters we find the RF frequency is higher than 918 MHz, so we bring it down by 4 MHz. In one we find the TCK period is below 195 ns so we increase it to 207 ns. In one the TCK period is correct but the mark-space ratio is only 46%. We fix the mark-space ratio. There are two transmitters marked 34.7.

We vary the transmit clock period of transmitter 34.7 and some others, and record reception for all transmitters after measuring their period. We obtain the following graph showing how reception varies with transmit clock period, provided that the mark-space ratio is between 48-52%. It looks like the period must lie within the range 195-210 ns for reliable reception. We make sure that all 15 of the A3028s lie within this range.


Figure: Reception versus Transmit Clock Period. We have two measurements for each A3028, marked First and Second, and we have one measurement for each A3019.

While testing the transmitters, we place each one on our spectrometer's Damped Loop Antenna (A3015C). We observe power at the transmitter's center frequency of −26 dBm. Our interference power peaks at −45 dBm, and we obtain good reception so long as interference is 10 dB less than our signal, so we are certain obtain reliable reception with the A3028 on the loop antenna.

2014

JAN-14

[07-JAN-14] We have 14 of A3028 encapsulated. This batch has un-stretched springs for leads, so they are more flexible, and the silicone coating is thinner. We turn them all off, put them in a box and shake them around together. Some of them turn on. We turn them off, place them apart on our bench and pick each one up and handle it and tap it. None of them turn on. We put them together in a box and some of them turn on. It appears that some of the batteries on these transmitters have become magnetic, so that transmitters can turn one another on and off.

We measure frequency response of all 14 transmitters. All are within 1 dB of nominal, and the pairs of channels match to within 0.2 dB. All turn on and off several times with a magnet. We leave them turned off soaking in water. There remains fifteenth transmitter, No36.1, that won't turn off. We place it in our isolation chamber to be part of our reception experiments. We will leave it running continuously from today to see how long it will last. It has already been running since 31-DEC-13.

[13-JAN-14] Our 14 transmitters have been soaking for 5 days. We also have No32.6, which has been soaking for several weeks. No34.9 has good frequency response but poor reception (75%). We measure its transmit clock period to be 190 ns, which is too low. No34.5 produces a square wave on its No6 input, which is its X input. All other transmitters: 32.6, 34.1, 34.3, 34.7, 34.11, 34.13, 35.3, 35.5, 35.7, 35.9, 35.11, 35.13, and 36.7 have perfect frequency response and good reception. We put No34.5 in the oven at 60°C.

We discover a bug in the P3028 firmware: the Y channel is the lower-number of the two channels, when our expectation was that it would be the higher-number of the two channels. We will leave things as they are for now, until we have shipped the 20 dual-channel transmitters ordered in Job 1141. Then we will correct the problem.

[14-JAN-14] No34.5 has good reception and perfect frequency response today, after a bake of several hours yesterday.

[17-JAN-14] After a few days in water, No34.5 has poor reception, and sensitivity of No6 channel to mains hum appears higher. No square wave.

[21-JAN-14] No34.5 is square waving again, and has poor reception. The square wave appears on both channels. We turn it off, let it sit for a few minutes in air, turn it on again, and it no longer generates a square wave, but reception is still poor.

FEB-14

[12-FEB-14] No36.1 has been running since 31-DEC-13 and is still going strong.

[21-FEB-14] Shown below is the thermoplastic over-molded A3028A with four coats of silicone, compared with our own epoxy-encapsulated A3028A with four coats. The silicone went on the thermoplastic very well, but we messed up the final two coats with dry air.


Figure: A3028A with Over-mold Version One, and Hand Epoxy Encapsulation.

The over-molded device has exterior dimensions approximately 15.0 mm × 15.0 mm × 9.0 mm = 2.2 ml, while the epoxy encapsulated device has exterior dimensions 14.0 mm × 14.0 mm × 9.0 mm = 1.8 ml. We would like to reduce the width of the over-mold by 1 mm if possible. We place the over-molded transmitter in water to measure its switching noise. Both channels have the same 3-μV amplitude switching noise at the exact same 20.75 Hz.

We take out transmitter No34.5 and find that it will no longer turn on.

[25-FEB-14] We examine the A302801B layout and find the most likely place at which moisture-invoked feedback from the output of the EEG amplifiers to their inputs could cause the square wave problem we observed in the A3019, and in No34.5. In the A302801B layout, the X amplifier is on the bottom side and the Y amplifier is on the top. The X input appears on pad U5-3, and ×100X on the CH0 via. These are separated by 1.4 mm. The Y input appears on U6-3, and ×100Y on the CH1 via. These are separated by 0.6 mm. Oscillation occurred with condensation in the A301901B layout because the input and output of the ×100 amplifier were separated by 0.6 mm. But in No34.5, X was oscillating, not Y. The ×100X appears on R11, which is 0.7 mm from the X input pad. If there were residual acid flux around the X and Y input pads, and combining with condensation, it would cause oscillations in X rather than Y. Another potential source of oscillation is the 0.5-mm separation between the X on U5-5 and ×40X on R8, and between Y on U6-3 and ×40Y on R15.

[27-FEB-14] We have a batch of thirteen A3028Es, rat-sized single-channel. Load batteries with no-clean water-soluble flux. Wash in hot water. Bake. One won't turn on, drawing 44 mA from the battery. Replace U9, re-calibrate RF center frequency and it works. Clip extension in preparation for programming and note clipping requires us to torque the board up from the battery and sends a shock through the circuit board.

New firmware P3028A03 provides the correct channel numbers for X and Y in the dual channel versions.

[28-FEB-14] Check thirteen A3028Es and all turn on and off, show mains hum.

MAR-14

[03-MAR-14] We take four broken A3019 and A3028 circuits, load BR1225 batteries, and coat them as they are with aerosol acrylic and aerosol silicone conformal coatings. We bake at 60°C for ten minutes, coat again, and bake for 20 minutes. We remove the batteries to see how well the interior components are coated. The coatings have a UV indicator, so we are able to take the picture below in our UV lamp to show where the coating is present.


Figure: Conformal Coating in UV Light. For an uncoated circuit in the same light see here.

The large square chip is the logic chip on the top side, beneath the battery. Its top surface is flush with the battery bottom surface when the battery is loaded. The silicone has penetrated between the two surfaces better than the acrylic. Close inspection of the circuits reveals that both coatings cover the circuit board and resistors below the battery. Upon close inspection of all boards, we note many bubbles and imperfections in the acrylic coating, and few imperfections in the silicone. We remove the MSOP-8 package on the top side and observe both acrylic and silicone have covered 90% of the area beneath the package, as shown below.


Figure: Penetration of Coating Under MSOP-8 Beneath Battery.

We load a battery onto an A3028A along with three leads and antenna. We apply three coats of silicone conformal coating. We baked for at least 5 minutes at 60°C between coats and for three hours afterwards. The silicone has coated the circuit board, and had penetrated beneath the larger components, but the corners of the P0402 resistors are bare. We connect an auxiliary battery, turn the transmitter on, and find we can taste the battery voltage on the bottom side of the board. We place the circuit in water and its X and Y channels oscillate together at about 1 Hz.

No36.1, an A3028D, has been running since 31-DEC-13 and is still going strong.

[06-MAR-14] We receive this xray image of the over-molded A3028A from one of our assembly companies. The picture below is a close-up of the space under the battery, with contrast and brightness enhanced to show the cavity beneath the battery wrapper.


Figure: X-Ray Image of Space Beneath Battery in Over-mold.

The space between the battery body and the top-side components is filled. The spaces between the battery tabs and the battery body are filled. If such wide, thin spaces are filled, we are hopeful that the spaces between P0402 components are also filled.

[10-MAR-14] Transmitter No36.1 is near the end of its life: battery voltage is 2.2 V. It has been running since 31-DEC-13, a total of 70 days, or 10 weeks, which is 2 weeks longer than we guarantee. We put the transmitter in water with the X and Y leads immersed and the C lead outside. We see high-frequency noise as shown below.


Figure: Radio-Frequency Pick-Up with C Out of Water.

This noise is the same as the high-frequency noise that arises on the A3019 when there is a break inside the insulation of the X− lead.

We have 13 transmitters No38.7 through 39.5 (missing 38.6) that have been soaking in water for four days. We test reception and frequency response with 20 MΩ source. All are normal except No39.1 which has poor reception and 38.9, 38.10, and 39.2, which have gain 3 dB below normal at 100 Hz. We connect these three to a 50-Ω source and frequency response for all three is normal. We put all of them in the oven to bake.

We suggest that the low gain described in the previous paragraph is due to a capacitance between the X input and the amplified X, location shown here. The gain is normal at 10 Hz because the capacitor's impedance is much greater than our 20-MΩ source impedance. The gain drops by 3 dB at 100 Hz because the capacitor loads the source with −10j MΩ in parallel with the X input's 10-MΩ input resistor. When we drive the input with a 50-Ω source, the condensation capacitor does not load the source, so the gain is normal at all frequencies.

[12-MAR-14] Our transmitters No38.7 through 39.5 (missing 38.6) have been baked for a few hours. Gain versus frequency is nominal for all 13 transmitters. Our Octal Data Receiver is malfunctioning, giving the impression of poor reception on all channels. We replace the data receiver and all 13 transmitters have perfect reception. We measure switching noise in water.


Figure: Noise on A3028E Single-Channel Rat Devices No38.7-39.5 (missing 38.6) in Water.

The switching noise is no more than 6 μV and the frequency is 22±1 Hz. The variation in frequency is five times smaller than for the A3019.

Transmitter 36.1 has battery voltage 2.17 V but is still running after 72 days. It detects mains hum, heart beat from two people, and has normal noise.

[19-MAR-14] Transmitter 36.1 has battery voltage 1.94 V but is still transmitting intermittently. It records mains hum.

[21-MAR-14] We have a batch of nine A3028Es, No39.6-39.14, with 150-mm un-stretched leads. We applied acrylic coating to the critical regions of the A302801B layout. We encapsulate with epoxy, touched up with nail polish, and applied five coats of silicone. The result is a transmitters with average body volume 2.8 ml (we immersed up to the antenna base and measured total displacement of water 24.8 ml). For the past year we have been applying eight coats to rat transmitters, but we now believe eight coats is excessive, because moisture problems do not arise from penetration of the silicone, but rather through condensation inside a sealed silicone coating. We leave in water to soak.

We have a batch of four A3028Bs, No40.1-No40.4, with 50-mm un-stretched leads. We apply acrylic coating, encapsulate in epoxy, coat five times with silicone. Two of these have a gold-plated pin for the X electrode (electrode type F). We leave in water to soak.

[24-MAR-14] Transmitters No39.6-40.4 have been soaking in water for three days. Frequency response of all thirteen transmitters is within 1 dB of nominal at all frequencies. Reception is perfect in our small faraday enclosure. We obtain this plot of switching noise in No39.6-40.4 (One vertical division is 0.8 μV).

We have a yield of 13 out of 13 after a three-day soak. We resolve to ship a batch of 8 A3028Es to ION and a batch of 3 A3028Bs to Edinburgh.

[26-MAR-14] We have completed and double-checked our A302801C layout. We describe the changes and give links to new files here.

[28-MAR-14] We have an A3028A that won't turn off. It's the one we received with over-mold from an assembly company, and it has been running since at least 12-MAR-14 when we first examined it and determined that it was stuck on. So far it has seen 16 days of continuous running, and our expected battery life for this device is 15 days. Battery voltage is 2.52 V. We drop it in water with the lead ends stripped and observe noise 37 counts rms on both channels.

[31-MAR-14] The above-mentioned over molded transmitter is no longer running. Operating life for the A3028A appears to be a little over 16 days.

APR-14

[11-APR-14] We assemble two A3028P single-channel transmitters for implantation in rat pups. At the moment, they have their programming extensions attached. They are equipped with freshly-charged PP031012AB 19-mA-hr lithium-ion batteries. We obtain this plot of A3028 noise when powered by a lithium-polymer battery, before encapsulation, when we expect no switching noise.

Noise amplitude in counts rms and battery voltage in Volts are as follows; No40.6: 19.4 and 4.3 V, No40.7: 17.5 and 4.3. For comparison we have an A3028A with a fresh, BR2330 battery with No13 34.0 and 3.0, No14 33.0 and 3.0. With battery voltage 4.3 V, one count is 650 nV, and for 3.0 V one count is 460 nV. We expect the noise amplitude to drop from 33 to 23 counts rms when we increase battery voltage from 3.0 to 4.3 V.

The center of the RF spectrum of both A3028Ps lies within 913-918 MHz. The transmit clock period lies within 195-215 ns. Reception in our faraday enclosure is perfect. Dynamic range at the X input is 43 mV. Gain and frequency response is nominal for both devices.

We find burrs on the battery wire solder joints. We re-solder the joints and wash the device with the battery still attached. Both batteries lose voltage. One appears to be discharged and accepts a re-charge. The other will not re-charge. We replace the batteries and epoxy them as before.

[14-APR-14] Transmitters No40.6 and 40.7 look good, but No40.6's battery is drained for reasons we cannot explain. We did not leave the enabling jumper in place, so the battery should have been isolated. We recharge both batteries and monitor their battery voltages through their X measurement with the Recorder Instrument and a 200-Ω resistor draining the battery directly.


Figure: Battery Drain with 200-Ω Resistor. The battery is a PP031012AB 19 mA-hr lithium-ion polymer cell.

The 200-Ω resistor load drains the battery almost the maximum recommended rate. By the time we end the experiment, No6's battery has delivered 12 mA-hr and is down to 2.5 V. The transmitter stops working. No7's battery has delivered 14 mA-hr and its battery voltage is still 3.5 V. The No6 battery has been damaged by over-discharge, but still provides more than half its nominal capacity.

We encapsulate two A3028Ps. One was faulty before, both were faulty after. The first one, No6, draws 800 uA when asleep and 900 uA when awake. Even when asleep, it will drain its battery in about twenty-four hours. This explains why it worked when we first assembled it, then the battery was completely drained after a day. When it is awake, it transmits fine.

The second one, No7, draws 2.0 uA, but when you turn it on it draws 170 mA, which is enough to generate heat, and it does not transmit. So we let it sit like that for a while and then feel the various parts It turns out to be the RF oscillator, U9, that's faulty. Further inspection reveals that the unused pad under pin U9-1 has shifted over and is touching U9-2. By this time we knocked off the BGA-5 chip, and can't get another on there because of all the silicone.

We conclude that both circuits were damaged during experiments and encapsulation. We should try again.

MAY-14

[05-MAY-14] We have a batch of ten A3028Es, No40.12-41.7, and two A3028Bs, No40.9-40.10. They have been soaking in water for five days, with three lots of hot water to provoke internal condensation. We measure gain versus frequency and find it to be within 1 dB of nominal for all transmitters. Switching noise peaks are all in the range 20-24 Hz, peak amplitude 5 μV, average amplitude 2 μV. No sign of spikes on No40.12, which has amplitude 5 μV. Thus we have all twelve transmitters functioning perfectly.

[09-MAY-14] We have two A3028P rat pup transmitters. We put them in hot water yesterday and left them soaking while turned off. Here they are after blow-drying, along with a non-functional model we assembled earlier. The model shows how we want the antenna and leads of the A3028P to be arranged, in a plane. The two working prototypes have their leads and antenna coming coming off the circuit board at odd angles.


Figure: Prototype A3028P Rat Pup Transmitters. Left No41.9, center No41.8, right a non-functioning model.

We hope to fix the direction of the leads in future assemblies. In water, we measured battery voltage and noise amplitude before soaking yesterday and obtained for No41.8 4.2 V and 9 μV and for No41.9 4.1 V and 8 μV. Today we obtain for No41.8 4.1 V and 7 μV and for No41.9 4.2 V and 10 μV. We measure frequency response and find it to be within 1 dB of nominal for both.

[12-MAY-14] No41.8 and 41.9 still transmit a strong signal. Analog gain is 30 dB too low. We backe both in the oven for half an hour and 41.8 recovers fully, showing correct gain with frequency, while 41.9 shows some improvement. Leave them in the oven.

[14-MAY-14] No41.8 and 41.9 now working well. Battery voltage is 4.1 V. Frequency response is within 1 dB of nominal.

[19-MAY-14] New batch of transmitters, No41.10-42.9 placed in hot water and allowed to cool and soak for four days. Today we measure frequency response. All are within 1 dB of nominal except for No42.2, which is 2 dB below nominal at 130 Hz, there being no bump in the response before the cut-off. We measure switching noise, and obtain this plot showing noise less than 5 μV and within the range 21-23 Hz.

[28-MAY-14] We receive a batch of 50 of our A3028 assemblies with A302801B circuit board from Advanced Assembly. Below is one of the x-ray images of the two BGA packages we received from them.


Figure: Xray Image of U3 and U8.

We program and calibrate ten circuits, and test the magnetic switch, which remains a BGA-5 in this version of the circuit board. All work fine. There is no sign of the problems we have had in the past with the BGA-5.

JUN-14

[09-JUN-14] We have two new A3028P pup transmitters No43.10 and 43.11. They have been soaking for four days in water. We used acrylic coating on the analog circuits prior to applying three coats of silicone by dipping. Batteries are sealed with epoxy then three coats of silicone by dipping. The antenna is 45 mm long and bent into a tighter loop.


Figure: Prototype A3028P. This is either No43.10 or No43.11.

In our faraday enclosure we obtain 100% reception from both transmitters when they sit on our antenna. We place No43.10 and an A3019D No29.12 on an A3015C Loop Antenna on our table. We place the transmit antenna loops right on the A3015C surface, so the batteries are on top. Reception is 100.0% for both. We place the battery side down. Now the A3028B batteris is between the receive and transmit antennas. Reception is 59.9% from the A3028P and 99.9% from the A3019D. The A3028P should be implanted with the battery facing the body and the antenna facing the skin.

We measure frequency response. Gain is within 1 dB of nominal from 0-250 Hz for both transmitters. We place in water to measure noise and battery voltage. For a few minutes, No43.10 shows transients up to full scale, with frequency a few Hertz. It settles down to an average of 28500 counts and 14 counts rms, so battery voltage is 4.1 V and noise is 9 μV. No43.11 shows no transients, battery voltage 4.1 and noise 10 μv.

We have fourteen A3028E transmitters 42.10-43.9. They have been soaking in water for ten days, including three hot-water soaks. No42.10 won't turn on. Frequency response of all remaining transmitters is within 1 dB of nominal. Switching noise is less than 3 μV for all. Transmitter No43.4 has a flaw in its blue lead. We have twelve that are ready to ship.

[13-JUN-14] We have 5 of A3028AV2 made with A302801C circuit boards. See here for photograph of top side. The two large holes are for the BR1225 battery. When we load the battery into these holes, it is centered perfectly on the square of the circuit board. Note barrier pads in top-right near X and Y inputs. We program four as dual-channel and one as single-channel transmitters. We use the new V4 firmware, which provides compiler directives to select A302801B and A302801C circuit boards. We measure gain versus frequency and find it to be within ±1 dB for all channels and frequencies.

[19-JUN-14] We receive an A3028AV2 on an A302801C circuit board with leads, antenna, and battery loaded, and new over-mold applied. See photograph below.


Figure: Over-mold Version Two. The circuit inside is the A3028AV2 made with A302801C circuit board, for which the battery is centered exactly by holes large enough for its tabs. Item 1 is a flange that will be absent in the production version. Item 2 is damage to the over-mold when it was removed from the mold.

The A3028AV2 over-mold is 7.7 mm high and 13.6 mm square. With two coats of silicone, the device will be 8.2 mm high and 14.1 mm high. Two coats may not provide a reliable seal over the irregularities in the mold surface. With three coats of silicone, we should get a good seal, but the device will be 8.4 mm high and 14.4 mm square. Our hand-encapsulated A3028A is not rectangular. Its height is varies from 8.0 to 8.4 mm and its width from 13.4 to 15.3 mm.

We program and calibrate the over-molded circuit. It works well with an external battery. It does not work with its internal battery. The internal battery voltage drops from 2.5 V to 1.0 V when we close the internal battery switch. The battery is exhausted. We connect another battery in parallel. After a few minutes, the internal battery voltage has risen to that of the external battery. Roughly 0.3 μA flows into the circuit. We close the internal battery switch. The external battery supplies 56 μA through R1 (2.7 V / 50 kΩ). We turn on the transmitter. The external battery supplies 187 μA (56 μ + operating current of dual-channel transmitter). We conclude that the internal battery is exhausted.

[20-JUN-14] At one of our assembly companies, they find that the batteries we sent them a month ago are all drained of charge. We sent the batteries in anti-static foam, which is electrically conducting.

We note that the circuit board entering the over-mold is at a slight angle to the horizontal, as a result of tension in the antenna and leads. The antenna and leads emerge vertically and must be bent to go horizontal. Instead of soldering the leads through the holes, we now plan to solder them as shown below.


Figure: Surface-Mount Leads.

The pads two which the leads are soldered are all secured by through-plated holes. The leads are secured by a rivet of solder.

[25-JUN-14] Transmitters 43.12-44.7 have been soaking in hot and cold water for a week. We measure frequency response, reception, and noise. All give gain within 1 dB of nominal for 0-250 Hz, reception 100% in faraday enclosure, and noise less than 40 counts rms after dropping in water. The plot below shows the switching noise in a 32-s interval.


Figure: Switching Noise in Batch 43-44. Fourier transform of 32-s interval. Frequency steps are 1/32 Hz (0.03125 Hz). One vertical division is 0.8 uV.

Peak switching noise is 6.4 μV. We see the second, third, and fourth harmonics of the switching noise clearly in the spectrum.

JUL-14

[02-JUL-14] We have A3028AV2 encapsulated in epoxy and silicone with no acrylic coating on the EEG amplifiers. We want to find out if the barrier pads of the A302801C layout, and its greater separation between the input and output components, will eliminate condensation faults without the acrylic coating. These five transmitters have been soaking in water for a week with four hot water charges. They are No45.1 (A3028D), No45.3 (A3028D), No45.5 (A3028A), No45.7 (A3028D), and No45.8 (A3028B). Thus all but the last one are dual-channel. We connect to our 20-MΩ sinusoidal source and find the gain of the dual-channel devices lies within 1 dB of nominal. The difference between channels is less than 0.2 dB. Device No45.8 has gain 1 dB higher than normal from 1-20 Hz, nominal gain 20-110 Hz, shows only a 1-dB increase in gain at 130 Hz instead of 3 dB, and has the correct cut-off frequency. We place No45.8 in parallel with one input of No45.5 and note that the two have the same response to within 0.2 dB up to 110 Hz, when No45.8 is 2 dB too low. None of the transmitters generate a square wave when left open-circuit.

[03-JUL-14] After twenty-four hour bake at 60°C, No45.8 has nominal gain from 1-110 Hz, shows only a 1-dB increase in gain at 130 Hz, and has the correct cut-off frequency. Yesterday, before baking, the gain was 1 dB higher than nominal from 1-20 Hz, which is within specification. The lack of a 3-dB bump in gain at 130 Hz is out of specification, but persists after baking. So far, these first five A3028AV2s, encapsulated without acrylic coating, show not sign of condensation problems. Peak switching noise for the five transmitters is 4 μV in the range 20-22 Hz with both channels of each dual transmitter agreeing about the amplitude of the switching noise.

We have A3028AV1 circuits encapsulated with acrylic, epoxy, and silicone, No44.8-44.14, No46.1-3, all A3028B single-channel mouse transmitters. These we made with wires soldered flat on the pads, as shown here. The result is leads that emerge in the correct direction without bending. Frequency response of all amplifiers within 1 dB of nominal. Switching noise 6 μV maximum. We measure volume by displacement of water and get 1.36 ml each. We measure by weighing in and out of water and get 1.45 ml. On average, around 1.4 ml. Mass is 2.7 g.

[31-JUL-14] We have 12 of A3028E-AB, No46.5-47.2 that have soaked in water for five days. We measure gain versus frequency, all are within ±1 dB of nominal. We measure reception, all are 100% in enclosure. We measure switching noise and all 20-22 Hz, 0-6 μV. All turn and off multiple times without error.

We have No45.11 A3028A with thermoplastic over-mold and one coat of silicone. The one coat leaves visible cavities around the electrode leads. We left it soaking in water for three days. Now its battery voltage is 2.0 V, reception is poor, and gain is 20 dB too low.

We have No48.1 A3028A with thermoplastic over-mold and three coats of silicone. There are no visible cavities in the coating. We left to soak for three days. Its battery voltage is now 2.7 V, gain versus frequency within 1 dB of nominal, and within 0.2 dB between channels. Switching noise is 1.2 μV at 19.5 Hz. Reception is good. Magnetic switch is working well.


Figure: Over-molded A3028A with Three Coats of Silicone.

The transmitter body fits in a rectangular volume 14.3 mm × 14.0 mm × 8.5 mm = 1.7 ml. The volume occupied by the transmitter is 1.6±0.1 ml. Our most compact hand-encapsulated A3028A has volume 1.4±0.1 ml. The three coats we applied to No48.1 increased its thickness from 7.8 mm to 8.5 mm. This accumulation agrees well with the 125 μm per-coat thickness specified in the MED10-6607 data sheet.

AUG-14

[13-AUG-14] We have 9 of A3028B, No40.8, 40.11, 47.3, 47.6, 47.7, 47.9, 47.10, 47.12 and 2 of A3028A, No47.4 and 47.13. They have been soaking for four days. All are within ±1 dB of nominal frequency response. All give 100.0% reception in enclosure. Switching noise is 8 μV for No47.4 (both channels the same). Others have less switching noise. Magnetic switches all respond well.

SEP-14

[03-SEP-14] We have 8 of A3028B-CC, No49.2, No49.5-No49.11. We note that 49.11 is labelled 49.3 so we apply colored silicone over the label to obscure it. All transmitters produce less than 8 μV of switching noise and have frequency response within 1 dB of nominal. Reception is 100.0% in a faraday enclosure.

[12-SEP-14] Transmitters A3028E-FB No42.8, 44.6, and 44.5 have failed at Philipps University, Marburg after eight weeks implanted. No42.8 works fine, except it keeps turning itself off. This circuit was one upon which we replaced U3 by hand. No44.6 and No55.4 have exhausted their batteries. Even if they were left on from the moment we shipped them, they should still be running. We suspect excessive current consumption, which is a problem we observed in some members of the batch of circuits from which these two were taken.

We have a new batch of 100 circuits in which U3 is a UDFN-6. They work fine, except two out of twenty so far have had a short between U3-1 and U3-6. The assembly house glued U3 in place, so we must remove it with pliers. We replace and the board works fine. We note that the spacing between U3-1 and U3-6 is too small. The U3-1 pad is too large. We create A302801D layout and reduce the pad from 19 mils to 16 mils.

[22-SEP-14] We have a batch of twelve transmitters No49.14-50.11. All have been soaking in water for three days. All have frequency response within 1 dB of nominal, perfect reception in a faraday enclosure, and switch noise less than 6 μV.

[23-SEP-14] We hear from Pishan at UCL that 44.10 had symptom "no signal" after implantation on 11-AUG-14. This transmitter was shipped on 03-JUL-14, so it is possible that, once it arrived at UCL, it turned on and has since exhausted its battery.

OCT-14

[01-OCT-14] We have transmitters A3028E-AB No50.14-51.13. The silicone has cured for two days, but we have not soaked in water at all. We must ship them today. All have gain within 1 dB of nominal, perfect reception. We ship all but No51.3 and No51.4, which stuck together during curing, so we must touch up the outer coat. Switch noise is less than 6 μV for all the transmitters we shipped.

[03-OCT-14] Transmitter 50.12 and 50.13 are encapsulated in acrylic conformal coating and silicone. We injected silicone under the battery twice and dipped four times. Battery voltage is 2.5 V for some reason, but frequency response is within 1 dB of nominal. We place in water. Two hours later, No50.13 still had nominal frequency response, but No50.12 is generating a 1-Hz full-scale square wave.

[06-OCT-14] Transmitter 50.13 has been sitting in water. Its gain is 20 dB too low when driven by a 20-MΩ source impedance. We connect it in parallel with another transmitter and the gain of both is now 20 dB too low. We drive 50.13 with a 50-Ω source and we obtain nominal gain at 10 Hz, 3 dB too low at 100 Hz, and 6 dB too low at 130 Hz. We place 50.12 and 50.13 in the oven at 60°C

[08-OCT-14] After 48 hours in the oven, 50.12 and 50.13 both perform perfectly.

[10-OCT-14] Put batch No51.14-52.13 in water. Kirsten tells us she is certain she forgot to put the acrylic coating on the circuit boards.

[14-OCT-14] Batch No51.14-52.13 still soaking in water, turned off. These are A3028V2 made with the A302801C circuit board. We check frequency response on all 14 transmitters and find it to be within 1 dB of nominal in all cases. We leave in hot water.

[15-OCT-14] We examine 50.12 and 50.13 closely. Both have many bubbles in the acrylic coating around the amplifier parts. We consider whether there is a reaction between the acrylic coating and directly-applied silicone dispersion. We apply two coats of acrylic to a bare circuit board, then coat with silicone and see no bubbles. We have observed such bubbles when acrylic coating is tacky due to drying out. Transmitters 50.12 and 50.13 were made with tacky acrylic. Each transmitter has a deep hole behind the battery terminal, leading to the interior space beneath the battery. We fill these holes with a syringe and leave to cure. We review our history of acrylic coating and condensation-related problems. We started applying acrylic in September 2013 with the A3019A/D transmitters. We saw an immediate reduction in the incidence of condensation problems after our four-day water soak. As we moved to water washing and complete acrylic coating, condensation problems disappeared. Thus we believe the acrylic coating solved the A3019 condensation problem. The A3028V2 introduces a new layout with barrier electrodes and more distance between EEG amplifier input and output. Both 50.12 and 50.13 are A3028V2s. We have had no condensation problems with this circuit until the 50.12 and 50.13. Our hypothesis now is that an imperfect acrylic coating (bubbles) combined with an imperfect silicone coating (holes) resulted in condensation problems even with the new A3028V2 layout.

[19-OCT-14] At ION, transmitter No13, we're not sure which batch, has failed after several days implanted with an ISL. Its last moments are shown here.

[20-OCT-14] Two more transmitters have failed at Philipps University, Marburg. The screen shot below shows the recording shortly before failure from transmitters No50.1 and No50.2.


Figure: Recordings from No50.1 and 50.2 Shortly Before Shut-Down. Eight-Second Interval 32-40 s. Green: No50.1. Blue: No50.2. Taken from archive M1413474679.ndf.

The average value of signal No50.2 in this interval, and more clearly in later intervals around 170 s, is around 20,000 counts, which implies a battery voltage of 6.0 V. The maximum possible voltage supplied by our lithium primary cell BR2330 is 3.0 V. When the average converted value is 30% of the full scale, the X input to the ADC (pin U7-2, see schematic) must be only 30% of the ADC power supply (VA). But the average value of this X input is 1.8 V when the amplifier is working correctly. Thus the amplifier is damaged.

The high frequency noise on No50.2 at 37-38 s has power from 75 Hz all the way up to 255 Hz. At time 192 s, there is full-scale noise from 0-255 Hz. The gain of the amplifier at 255 Hz is only 1, so a full-scale 255 Hz implies a 2.7-V 255-Hz signal at the X input electrode. Assuming such an input is impossible, the amplifier must be generating this signal itself, which implies it is damaged.

The Philipps University experiment involves stimulating the brain with 20-V pulses, 40 ms apart, 0.1 ms duration, at 2 Hz, as well as more sustained stimuli every minute or two. The trace below shows two sets of pulses.


Figure: Electrical Stimuli as Seen On X Input. Scale is 0.4 mV/div vertical and 0.1 s/div horizontal. Taken from archive M1405508254.ndf, 81-82 s.

The pulses are around 2 mV peak to peak. The more sustained stimuli reach an amplitude of 4 mV. Looking at these traces, we see no reason to suppose that the input to our transmitter is being driven by a signal outside the absolute maximum ratings of op-amp U5. The inputs are clamped by diodes inside the op-amp to its power supply voltages. Nevertheless, if we were to apply 20 V through 10 kΩ to the X input, we would see around 17 mA flowing through these clamping diodes, which exceeds their maximum rating of 10 mA. Thus it is possible to damage the transmitter with a 20-V stimulus delivered through brain tissue. Damaged op-amps can develop erratic input offset voltages and consume excessive current. If U5 starts to consume 1 mA, VA will drop to 1.7 V and the ADC will stop working. It will produce only 0's or 1's. We see such intervals in the No50.2 recording. Furthermore, the battery will drain in about ten days.

[20-OCT-14] After two more injections of silicone, making six in all to fill the holes behind the battery terminals, we are satisfied that No50.12 and 50.13 are sealed, so we put them in water.

[20-OCT-14] Batch No51.14-52.13 has soaking since 10-OCT-14. This is the batch with no acrylic coating. Gain versus frequency is within 1 dB of nominal for all transmitters.

[21-OCT-14] We have No44.10 back from UCL. We shipped this device on 03-JUL-14 and Pishan implanted on 11-AUG-14, at which point she discovered that it would not turn on. We find the exterior silicone in perfect condition, the epoxy in perfect condition, and the leads also. we cut away enamel and epoxy to get to the battery terminals. The battery voltage is 0.13 V. We apply 3.6 V and 600 μA flows into the dead battery. We can turn on and off the transmitter easily. The analog input shows mains hum. We drop the applied voltage to 2.7 V. After a few minutes, off-state quiescent current drops to 7 μA. We turn on and current increases to 85 μA. This transmitter, when calibrated, had on-state current 79 μA and off-state current 2 μA. If we assume that we still have 2 μA flowing into the circuit in the off-state, then 5 μA is flowing into the battery, which means the on-state current is now 80 μA, which is very close to the original 79 μA. Our best guess as to why this transmitter failed is that it was left on and drained its battery.

[21-OCT-14] Silicone-encapsulated No50.12 and No50.13 have been in water for 24 hours. No50.13 shows −6 dB gain versus frequency, but normal bump and cut-off frequencies. No50.12 won't turn on. We strip silicone from the underside and measure battery voltage 2.6 V. Now it turns on. Gain is correct at 1 Hz and drops off rapidly above that, with random drifting baseline in the signal.

[21-OCT-14] We receive this diagram of the electrode arrangement at Philipps University, Marburg.

[20-OCT-14] Batch No51.14-52.13 has been in water since 10-OCT-14. Gain versus frequency remains within 1 dB of nominal. We switch cold for hot water. We measure switching noise and perform harmony test. All transmitters appear to be perfect. We leave them switched on and in water.

[22-OCT-14] Batch No51.14-52.13 all running, noise is normal. Average voltages are between 69% and 71% of full scale.

[22-OCT-14] Batch No52.14-53.13 has been sitting in water for 5 days. Noise is normal. Reception is perfect. Gain versus frequency within 1 dB of nominal except for 53.7, which is 1 dB below nominal at 140 Hz. Magnetic switches all reliable.

[24-OCT-14] Batch No51.14-52.13 has been running in water since 20-OCT-14 and soaking since 10-OCT-14. Noise is normal. Average voltage is 70% of full scale. Frequency response within 1 dB of nominal.

[27-OCT-14] Batch No51.14-52.13 has been running in water since 20-OCT-14 and soaking since 10-OCT-14. Average voltage is 70% of full scale except for 52.1, which is at 84% of full scale, with occasional steps lower or higher. The battery voltage appears to be around 2.1 V. When we apply a 10-mVpp input, we see saturation and inversion of the positive cycles at the top end of the amplifier's dynamic range. We reduce the amplitude to 3 mVpp and observe frequency response within 1 dB of nominal. Reception is 100.0% within our faraday enclosure. All others show perfect reception, normal switch noise, and frequency response within 1 dB of nominal.

No52.1 We remove silicone and release the +ve battery tab. We measure 2.27 V between the circuit VB pad and the positive battery tab. When we connect with an ammeter, we observe 1.6 μA in the inactive state and 7 mA when we switch on. At 7 mA, the battery would last only 36 hours, but 52.1 ran for four days with no change in battery voltage. We burn off epoxy over the -ve battery hole and connect 2.7 V. Inactive current is 1.9 μA, active current fluctuates between 12-15 mA. We connect 2.7 V directly to VD (using R4). Current is 13 mA. We activate and deactivate the magnetic switch. Current remains unchanged. After a few minutes, current is 9 mA. We increase the supply to 3.9 V and current increases to 20 mA, then drops over a few minutes to 15 mA. We remove U1 to make sure we have no current going back to the magnetic switch. Current is 16 mA at 3.1 V. We remove R4 and current is 14 mA at 3.1 V. Signal output is stuck at 0, but reception is perfect. We observe bursts of 910-MHz power every 2 ms in the RFPM Instrument. We compare spectrometer plot with an encapsulated transmitter and find peak power to be within a few dB. We remove U9. We now have no RF output, but current is 18 mA at 2.7 V. Remove U4. Current remains 18 mA at 2.7 V. We remove C3. Current drops to 20 μA, the correct quiescent current of U7 if supplied with 1.8 V. We are holding C3, preparing to measure its resistance and capacitance, when it shoots out of the tweezer tip and vanishes.

[28-OCT-14] Transmitter 50.5 has failed by discharging its battery while implanted at Philipps University. Looking at their recordings, it appears that a No7 transmitter is about to fail also.

According to the engineers at AVX, excessive leakage current in capacitors is caused by internal cracks, but its manifestation can be delayed by several months. The capacitor can be cracked during pick and placement at the assembly house, during depanelization, and during subsequent soldering near the capacitor. According to Ikeo et al. (Failure Mechanisms that Cause High Electrical Leakage in Multilayer Ceramic Capacitors), ceramic capacitors can fail by degeneration of their insulation layers resulting their behaving like resistors of a few hundred Ohms. The cause of degeneration they investigated is a voltage-driven chemical reaction assisted by the penetration of water and chloride ions into the capacitor through cracks or microscopic pores.

When we solder the antenna onto the A3028 circuit board, we use zinc chloride flux (acid flux) and high temperature. Capacitor C3 is the power supply decoupling capacitor for the radio-frequency oscillator. We give the location of this capacitor for various transmitters in the following table.

Version Capacitor Distance to
Pad (mm)
Distance to
Board Edge (mm)
A3019C7, 1nF0.36 top side0.33
A3028V1C3, 1nF1.6 bottom side2.4
A3028V2C3, 1nF1.0 bottom side0.41
A3028R1C3, 1nF0.58 bottom side2.1
Table: RF Oscillator 2.7-V Power Supply Decoupling Capacitor Location. We solder the antenna to the top side.

We shipped hundreds of A3019s and never observed this sudden discharge problem, even though we soldered with acid flux directly on the circuit board next to the capacitor, and we broke the circuit boards apart from one another by hand. This suggests that the problems with the capacitors on our new circuit boards is not with our soldering procedure, but either at the manufacturer or at the assembly house. We shipped a hundred A3028V1s and may have observed one such failure at ION recently (No13 see 19-OCT-14). The A3028V2 assembly we have been shipping since late September has shown 1 such failure out of 14 in our office, and at least 3 such failures out of 14 at Philipps University. Transmitter 50.7 may be failing the same way as we write. Transmitter 50.2 showed full-range oscillations at hundreds of Hertz, which could be caused by a leaky capacitor for C11.

We recently started cleaning the boards with a finer brush, which may be getting under the capacitors and cracking them. We will stop using this brush. We will examine all 1-nF capacitors on our A3028V2 boards for cracks. We will pre-tin the steel leads, wash them, and solder them with normal flux to the circuit boards. This will reduce the amount of chloride near the decoupling capacitor. But we suspect that we will see no cracks, because the cracks are internal, and that the damage was done during assembly, before the boards ever arrived at our office. Thus we will replace C3 on all boards.

[29-OCT-14] Transmitter 50.7 failed today at Marburg. Of the batch of 12 we sent them, 4 have failed by sudden battery drain and 1 by something crazy in the EEG amplifier.

We inspect the 1-nF capacitors on the bottom side of half a dozen A3028V2 circuit boards. These are C3, C9, C10, and C11. The figure below shows C9 and C10 on the circuit board with assembly company serial number ending in 049.



Figure: Two 1-nF Capacitors. Top Left: C9 seen from the side. Top Right: C10 seen from the side. Bottom Left: C9 seem from the top. Bottom Right: C10 seen from the top. Click on images to enlarge.

Looking closely at the top side of C9, we see a chip on the edge and a semi-circular mark. This mark is, we believe, the outline of the vacuum pick-up bit used to place the capacitor during machine assembly (see Figure 7 here). When too much force is exerted by the bit, such marks appear on top, and cracks occur on the bottom. Any crack in a capacitor is a means by which moisture can penetrate and, when voltage is applied, cause degeneration of the insulation.

Meanwhile, batch No51.14-52.13 has ben running in water since 20-OCT-14 and soaking since 10-OCT-14. No52.1 drained its battery two days ago. No further failures have occurred.

[30-OCT-14] Batch No51.14-52.13 has been running in water since 20-OCT-14 and soaking since 10-OCT-14. Thirteen of them remain working. Battery voltage is 2.66 V on average with standard deviation 16 mV. Frequency response is within 1 dB of nominal for all devices except 52.6, which has gain 1.5 dB above nominal at 140 Hz. We place transmitters 45.1, 45.3, 49.12, 49.13, 51.3, 51.4, 52.14, 53.13 in water, all turned on. We checked some of their frequency responses, and all of their battery voltages. We turn on acrylic and silicone encapsulated 50.3. Its gain versus frequency is once again with in 1 dB of nominal. We leave it running in water also.

We place a transmitter on a horizontal antenna in a faraday enclosure. We measure the power picked up by the antenna in four different orientations of the antenna. We do this with two mouse transmitters and receive up to −44 dBm from one and −40 dBm from the other. We repeat with three rat transmitters and receive a maximum of −42 dBm and −40 dBm from each. We perform the same experiment, but the transmitter is in a jar of water resting on the antenna. We receive up to −42 dBm from a mouse transmitter and up to −40 dBm with a rat transmitter. We placed the transmitters in a small petri dish of water. We obtained up to −36 dBm from a rat transmitter and −38 dBm from a mouse transmitter.

NOV-14

[03-NOV-14] Transmitters No51.14, 52.2-52.13 have been running in water since 20-OCT-14 and soaking since 10-OCT-14. Transmitter No52.6 generates its own sinusoid of amplitude 13,000 counts at 100 Hz when its input is driven by 0V through 20 MΩ. With inputs open-circuit, frequency drops to 80 Hz and amplitude increases to 22,000. When driven by a 10-mVpp, 50-Ω sinusoid, gain is 6 dB too high, but the shape of the gain versus frequency is within 1 dB of nominal. When connected in parallel with another 51.14 to 60 mVp-p through 20 MΩ, the gain versus frequency of the two transmitters is identical and correct. On its own again, with 20 mVp-p through 20 MΩ and we no longer see a square wave, but gain is 10 dB too high at 100 Hz. Place in parallel with No52.7 and gain versus frequency for the two is identical but 10 dB too high at 120 Hz. We put No51.14 in parallel again with 60 mVp-p through 20 MΩ and get identical gain, but gain is 7 dB too high at 120 Hz. Reception for No52.6 is perfect and noise is normal when in water. This transmitter has the square wave problem, and it appears to be varying in its severity as we perform our experiments. Transmitters No51.14, 52.2-52.5, 52.7-52.13 all have gain within 1 dB of nominal, perfect reception, and normal noise. In water, average signal is between 66% and 69% of full scale, indicating battery voltages 2.60-2.73 V.

Transmitters 45.1, 45.3, 49.12, 49.13, 51.3, 51.4, 52.14, 53.13 have been running in water since 30-OCT-14. Transmitters 45.1 and 45.3 are both dual channel. The figure below shows the frequency response of both channels in parallel for 45.1.


Figure: Gain Versus Frequency for No45.1, Both Channels. We apply a logarithmic sweep from 1-500 Hz in 16 s.

For similar plots for other transmitters see No45.3 (dual channel also), and No49.12, No49.13, No51.3, No51.4, No52.14, No53.13. All are within 1 dB of nominal, with No49.12 having gain 1dB above nominal at 120 Hz. Transmitter No50.13, which is acrylic and silicone coating, has gain 14 dB below nominal at 1 Hz, and 20 dB below nominal at 120 Hz.

Batch 54.1-54.10 has been running in water for three days. Average signal values 67-68% of full scale. The same is true for dual-channel 54.13 and 55.1. We measure gain versus frequency for the dual channel transmitters and find it to be within 1 dB of nominal.

[06-NOV-14] After baking, transmitter No52.6 shows gain versus frequency within 1 dB of nominal except at 120 Hz, where it is 1.5 dB above nominal. We set this one aside as a demonstration transmitter.

[07-NOV-14] After a total of five days running in water, we ship transmitters 52.14, 53.13, 54.3, and 54.4 to ION for job 1161.

[10-NOV-14] We have transmitters 49.12-13, 51.3-4, 54.1-2, 54.5-10, 55.1 (two-channel) running in water for a total of 4 + 5 = 9 days. Average signal values are all 67-69% full scale, except for 51.3, which is at 73%, indicating a battery voltage of 2.45 V.

No51.3 We check gain versus frequency for 51.3 and find it within 1 dB of nominal. Battery voltage remains 2.45 V. Silicone encapsulation looks intact all around. No sign of condensation inside. We remove silicone and disconnect positive battery terminal. We measure inactive current 1 μA and active current of 550 μA. We remove C3. On current is now 600 μA. We remove C5 and on current drops to 80 μ. We measure the capacitance of C5, it is 9 μF. Insulation resistance is greater than 20 MΩ. According to here, "Failed capacitors frequently recovered their insulation resistance at high temperature (above +200°C)." When we removed C5 with a soldering iron, we heated it to 400°C.

We hear from ION that 53.3 and 53.6 have failed after three days implanted, following a 5-day active soak. We do not yet know the nature of the failure. We place 49.12-13, 51.4, 54.1-2, 54.5-10, 55.1 in a jar of water in the oven at 60°C. All are running.

We receive from Philipps transmitters 43.3, 43.7, 50.1, 50.2, 50.5, and 50.9, all A3028Es. No43.3 Antenna is cut, red lead cut short. Reception 100%, picking up mains hum, average value with mains hum is 40,000. Gain versus frequency within 1 dB of nominal. Severe discoloration of purple enamel. No43.7 Reception 100%, picking up mains hum, average value 43,000. Gain versus frequency within 1 dB of nominal. Severe discoloration of purple enamel. No50.1 Reception 0%. Large cut across silicone on battery side. Open up encapsulation. Battery voltage is 2.1 V. Apply external 2.6 V. Inactive current 2 μA, on current 80 μA. Reception 100%, picks up mains hum. No50.2 Reception 0%. Open up encapsulation. Battery voltage 2.4 V and falling. Active current 2 μA, inactive current 80 μA. Picks up mains hum. Average signal 41.700 with 2.8 V applied. No50.5 Reception 100%, picking up mains hum, average value 51,000, gain versus frequency within 1 dB of nominal accounting for low battery voltage. Open up encapsulation. Battery voltage 2.4 V steady. Inactive current 2 μA, on current 80 μA. No50.9 Reception 0%. Open up encapsulation. Battery voltage 2.4 V and falling. Inactive current 2 μA, on current 7 mA. Reception when on is 100% and picks up mains hum. Heat up and then remove C3, C5, and C4. When we remove C4, on current drops to 80 μA.

[11-NOV-14] The average battery voltage today for 49.12-13, 51.4, 54.1-2, 54.5-10, 55.1, after 14 hr at 60°C running in water, is 2.83 V with standard deviation 0.025 V. Later in the day, after 24 hr at 60°C, No49.13 average value is at 90% of full scale, and No54.2 is at 73% of full scale. The others are still normal.

[12-NOV-14] Transmitters 49.13 and 54.2 are both draining their batteries. No49.13 Reception 100%. Open up encapsulation. Battery voltage 2.2 V. Inactive current 2 μA, on current 80 μA. No54.2 Reception 100%. Open up encapsulation. Battery voltage 2.4 V. Inactive current 550 μA, on current 600 μA. Remove C6, C5, C2. After removing C2, inactive current is 2 μA and on current is 80 μA. Later in the day, 54.6 is at 64% while the others remain at 68%.

[13-NOV-14] No54.6 does not transmit. Transmitters 49.12, 51.4, 54.1-2, 54.5, 54.7-10, and 55.1 all have average signal 67-69% of full scale. They have been running in water at 60° for a little over three days. We turn them off, dry them, and put them in the oven to recover. The black enamel of No54.6 has a pale discoloration. We open up the encapsulation. Battery voltage is 2.4 V. Inactive current 2 μA, on current is 1.5 mA. We heat up C5 with a lump of solder, but do not remove it. On current drops to 80 μA.

[21-NOV-14] We have batch E55.5, E55.6, B55.10, B55.11, A55.13, E56.1, E56.2, and D56.3 (the letter gives the version). We encapsulate without acrylic coating. We use DP270, a black potting epoxy with 60-minute work life. We coat four times with silicone. We turn them off and place in water on the morning of 19-NOV-14. Today we measure frequency response. All are within 1 dB of nominal, except for channels D56.3X, E55.6, E56.1, A55.13X, which have peak gain 1 dB higher than nominal. We put them back in water.

[25-NOV-14] We have batch R60.1-14 encapsulated with acrylic, EP965L epoxy, and four coats of silicone. We measure gain versus frequency and find it within 1 dB of nominal except for R60.1 and R60.2, which have gain 1 dB too low at 140 Hz. Reception is 100% for all transmitters. We calibrate our spectrometer with 910 MHz and measure center frequencies for R60.1-14 and obtain 920, 920, 920, 917, 923, 917, 919, 916, 917, 919, 916, 918, 918, 920 MHz respectively. Transmitter R60.5 is the one at 923 MHz. Switching noise is less than 4 μV for all devices. We place in water in the oven at 60°C at 9:00 am.

[26-NOV-14] Batch R60.1-14 have been running in water at 60°C for 36 hours. We replace 60°C water with 20°C water. Average signal values are 64-66% of full scale. Gain versus frequency within 1 dB of nominal. Reception excellent from all 14 simultaneously on one antenna. Switching noise normal. We apply three more coats of silicone to R60.5. The result is unattractive. We return all of them to the oven.

Batch E55.5, E55.6, B55.10, B55.11, A55.13, E56.1, E56.2, and D56.3 has been soaking in water at 20°C for one week. We measure frequency response. All are within 1 dB of nominal. We note that this batch has no acrylic coating. We turn them off and put them in hot water.

[28-NOV-14] Batch R60.1-14 have been running in water at 60°C for 72 hours. We cool them down. Battery voltages are normal, gain versus frequency is within 1 dB of nominal for all fourteen transmitters, noise is normal. We turn them off and put them back in the oven in a tray to dry out.

DEC-14

[03-DEC-14] Batch E55.5, E55.6, B55.10, B55.11, A55.13, E56.1, E56.2, and D56.3 has been soaking in water at 20°C for two weeks. Frequency response is within ± 1 dB of nominal, except D56.3X, E55.6, and A55.13X, which have peak gain 1 dB higher than nominal. Average values 44-45k, except A55.13 which is at 47k. We turn them off and put them in hot water.

[05-DEC-14] Rob Wykes at ION turned on Batch 52 and put them in water at 60°C for three days. At the end, he made the following observations.


Figure: Effect of 60°C Soak in Water While On, Batch 52.

We advise Rob that all four of the transmitters that were inactive upon removal are most likely faulty. This brings to 23 the total number of A3028E-ABs we must replace for ION.

[05-DEC-14] We have Batch R61.1-14 of A3028R-AB devices encapsulated with acrylic, epoxy, and silicone. We turn them all on and poach them for two hours at 60°C. Average signal value is 64-66% of full scale. We return them to the oven, running in water at 60°C.

[08-DEC-14] Batch R61.1-14 has been running in water at 60°C for three days. We put them in cold water. Average signal value is 64-66% of full scale, except 61.14 which is 69% of full scale. Switching noise less than 6 μV. Gain versus frequency within 1 dB of nominal for all devices.

[11-DEC-14] Batch E55.5, E55.6, B55.10, B55.11, A55.13, E56.1, E56.2, and D56.3 has been soaking in water at 20°C for three weeks. Frequency response is within ± 1 dB of nominal, except D56.3X, E55.6, and A55.13X, which have peak gain 1 dB higher than nominal.

[15-DEC-14] Batch R62.1-14 has been running in water at 60°C for four days. We checked that they were all transmitting and that switch noise was normal after a few hours in water, but did not measure frequency response before the poach. Today we find that all transmitters are still running. We get 100% reception from all of them. R62.10 has gain 1 dB too low at 120 Hz. R62.6 has gain 3 dB too low at 120 Hz. R62.12 transmits all zeros. We bake these three for an hour. The symptoms of R62.10 and R62.12 remain unchanged. But the gain of R62.6 is now within 1 dB of nominal. We ship all but R62.6 and R62.12. We put R62.6 in water to soak at room temperature.

[18-DEC-14] We take this picture of batch B63.1-14. Reception for all transmitters is 100.0% in our faraday enclosure. Frequency response is within 1 dB of nominal. We turn them all on and put them in water at 60°C.

[19-DEC-14] All transmitters B63.1-14 are still running. Average signal value between 45500 and 46000 (69.4-70.2% of full scale). Noise is on average 22 counts rms (9 μV). We turn them off and put them back in water at 60°C.

[22-DEC-14] Transmitters B63.1-14 have spent a total of four days in water at 60°C. We measure frequency response. All are within 1 dB of nominal except B63.3, which is 1 dB above nominal at 120 Hz. Here is a typical Neuroarchiver display of the sweep for B63.11. Transmitter B63.1 generates its own 78 Hz 3.2-mV oscillation when its inputs are connected by 20 MΩ. When we apply 30-mVpp sweep through 20 MΩ, we see this oscillation on top of the sweep input. When we apply 10 mVpp through 50 Ω, gain is within 1 dB of nominal. We connect B63.1 and B63.3 in parallel to 20 MΩ and observe 76 Hz 1.0 mV on both inputs, in phase and identical in shape. Average signal value for B63.1 is 70% of full scale, and for all others 66-68% of full scale. When in water, noise is less than 10 μV, with the fundamental of the switching noise less than 7 μV for all of B63.1-14. We turn off and place in the oven at 60°C in air. After a one-hour bake, B63.1 has nominal frequency response with 20 MΩ source and no sign of oscillation.

2015

JAN-15

[01-JAN-15] We have a batch of A3028R-AB for ION, numbers R64.2, 4, 8-14. All have gain versus frequency within 1 dB of nominal. Reception is perfect. Noise is 19-23 counts rms. We turn them all on and put them in water at 60°C. We have a batch of three A3028A-FFC dual-channel mouse transmitters we made by mistake. They were supposed to be A3028F-FFC. They are A64.3, A64.5, and A64.7. All have gain versus frequency within 1 dB of nominal. We turn them off and put them in water at 60°C.

Transmitter R62.6 has been soaking in water at room temperature for two weeks, turned off, following earlier reversal of condensation problems. We see the same condensation problem as before: gain is 3 dB too low from 10-150 Hz. We put in the oven to dry out.

Batch E55.5, E55.6, B55.10, B55.11, A55.13, E56.1, E56.2, and D56.3 has been soaking in water at room temperature, turned off, since 21-NOV-14. These were encapsulated with DP270 epoxy and silicone, no acrylic. Gain versus frequency is within 1 dB of nominal for all ten inputs, except for E55.6 and A55.13X, which have peak gain 1 dB higher than nominal. Reception is perfect.

We prepare firmware P3028A05, which accepts a version number to set the sample rate and enable one or both channels. We test the A3028F version. Current consumption is 255 μA, slightly lower than the current predicted by our current consumption formula.

[05-JAN-15] We remove A3028R-AB for ION, numbers R64.2, 4, 8-14 from the oven, where they have been running in water at 60°C for four days. R64.4, R64.8-14 have gain versus frequency within 1 dB of nominal and perfect reception. R64.2 has perfect reception but transmits only zeros. Transmitters A64.3, A64.5, and A64.7 have been turned off and poaching at 60°C for four days, although A64.7 is running when we put the transmitters in a pile on our work bench. Gain versus frequency within 1 dB of nominal, reception perfect.

[07-JAN-15] Transmitter R64.2 has been in the oven turned off an dry for two days. We take it out and it won't turn on.

[08-JAN-15] We have A3028F-FFC transmitters F66.1, F66.3, F66.5, and F66.7. These are two-channel transmitters each channel 1024 SPS and nominal bandwidth 320 Hz after decreasing the low-pass filter capacitors from 1000 pF to 510 pF. The following figure gives the frequency response of channel F66.1X.


Figure: A3028F Frequency Response. This is F66.1X.

We also have the same measurement for F66.1Y, and a normalized dual-channel plot for F66.5XY. The frequency response of all eight channels provided by the four transmitters is within 1 dB of nominal.

[09-JAN-15] We have batch R65.1-14, A3028R-FB. We turned them on and poached them at 60°C for 24 hours. Today all of them are running well, frequency response within 1 dB of nominal, switching noise 5 μV or less.

[12-JAN-15] Transmitter A64.1 has been soaking in water at room temperature, turned off, for three days. Gain on A64.1X is within 1 dB of nominal. Gain on A64.2X is 1 dB too high at 120 Hz, and when we connect A64.1X in parallel with it, the gain of A64.1X is also 1 dB too high at 120 Hz. This transmitter was encapsulated with an acrylic coating. We place A64.1 in the oven, turned off and dry, for one hour. The gain versus frequency is still 1 dB too high at 120 Hz.

We have batch R65.1-14 that have been poaching in water at 60°C for four days, turned on. Switching noise is normal for all. Reception is perfect for all. Gain is within 1 dB of nominal for all, except for R65.8 (1.3 dB too low at 120 Hz). We place R65.8 in the oven, turned off and dry, for one hour. The gain versus frequency is still 1.3 dB too low at 120 Hz. We go back and check our recording of batch R65.8 from 09-JAN-15, and it was 1.3 dB too low at 120 Hz then as well, but we did not note it at the time.

[13-JAN-15] We have batch E66.9-E66.14, E67.1-E67.7. Gain versus frequency is within 1.5 dB of nominal for all transmitters. Reception perfect. Battery voltages normal. E67.4 and E67.5 have their labels interchanged. Turn them all on and put them in the oven to poach.

We have F66.1, F66.3, F66.5, and F66.7, dual-channel 1024 SPS. Gain versus frequency within 1.5 dB of nominal. Battery voltages appear to be only 2.4 V. The current consumption of these devices is are 249 μA, 254 μA, 256 μA, and 240 μA respectively. This is like loading the battery with around 10 kΩ, and according to the battery data sheet, such a load will drop the output voltage by a few hundred millivolts. Leave in hot water on bench, turned off.

[13-JAN-15] We take batch E66.9-E66.14, E67.1-E67.7 out of the oven, transfer to warm water. We use a Toolmaker script to measure reception, battery voltage, and noise for all transmitters in water in our small faraday enclosure. Reception is good for all. Noise is around 10 μV for all, except E67.2, which is 22 μV. Battery voltage is 2.75 V for all except E66.14, which is at 2.68 V.

[16-JAN-15] Transmitters E66.9-E66.14, E67.1-E67.7 have been running for three days at 60°C. We take them out and measure gain versus frequency. All are within 1.5 dB of nominal, and within 0.5 dB of measurements before poaching, except for E66.14, which shows decreasing gain above 20 Hz. We place E66.14 in the oven to dry out. After half an hour, it still behaves the same, so we put it back for the weekend. We put new labels on E67.4 and E67.5 and cover with silicone.

Transmitters F66.1, F66.3, F66.5, and F66.7 have been soaking in water for three days. Gain versus frequency of all eight channels is within 1.5 dB of nominal.

[21-JAN-15] After four days baking, E66.14 has recovered.

[30-JAN-15] We have batch B67.8-14, B68.1-4. They have been running for one day in water at 60°C. Gain versus frequency is within 1.5 dB of nominal and recorded in one continuous archive.

FEB-15

[03-FEB-15] We have batch R68.5-12, R69.1-3. They have been running in water at 60°C for five days. Battery voltage is 2.7-2.8 V, switching noise is normal, reception is perfect, and gain is within 2 dB of nominal for R68.7-12, R69.1, and R69.3. R68.5 has battery voltage 2.1 V but is otherwise okay. R68.6 gives us only 90% reception in the faraday cage, but is otherwise okay. Its RF center frequency is 923 MHz. R69.2 has low gain. Gain measurements are in archive. We put R68.5-6 and R69.2 in the oven to dry out.

[04-FEB-15] After several hours of baking and several hours sitting in an oven that was turned off, R68.6 is still warm to the touch and has center frequency 920 MHz and we get 95% reception or Antenna No3 and 100% from Antenna No1. R69.2 gain is within 2 dB of nominal. No68.5 battery voltage is 2.6 V today. Frequency response is normal. But reception drops to below 10% for a few seconds at random. The spectrometer tells us that the transmit signal is 8 dB weaker than for R68.6. We turn it off and it turns itself back on again.

[17-FEB-15] We have batch B68.14, B69.5-13. We turn them on and poach them at 60°C for 24 hours. We remove, transfer to cold water, and check noise and battery voltage. Noise is less than 25 μV and battery voltage is within the range 2.55-2.65 V. Frequency response is within 2 dB of nominal. All switch on and off easily.

[23-FEB-15] We have batch R69.4, R69.14, R70.1-14 after three-day poach at 60°C. When we take the transmitters out of water, R70.5 and R70.11 were not running. All have switching noise less than 6 μV and total noise less than 30 μV, battery voltage within 2.6-2.7 V. Gain is within 2 dB of nominal for all, and reception is perfect for all. We put R70.5 and R70.11 back in to poach again, not knowing if they were off at the start of the last poach, or turned off of themselves.

[26-FEB-15] We have R70.5 and R70.11 after another three-day poach at 60°C. They are still running. Frequency response within 2 dB of nominal.

MAR-15

[02-MAR-15] Recent reports of failures in the field: R60.13 implanted 21-JAN-15 VBAT = 2.5 V on 21-FEB-15 (M1424569407.ndf), 200-μV step artifacts on 22-FEB-15 with VBAT = 2.4 V (M1424605404.ndf), 23-FEB-15 VBAT = 2.2 V with EEG being recorded (M1424688199.ndf), a few hours later we see the following final moments of the transmitter, with the apparent battery voltage dropping to 2.0 V (M1424691799.ndf). At ION we hear reports of six failures, but we have no details: R61.8 failed after 6 weeks implanted, R61.9 failed after 10 days, D45.1 failed after 6 weeks, R64.1 would not turn on before implantation, two more from batch R64 failed after three weeks implanted.

We have batch B71.1-12 after four-day inactive soak. Noise is less than 12 μV, switching noise less than 6 μV, reception is perfect in faraday enclosure, gain versus frequency within 2 dB of nominal except B71.2, which has gain +2dB at 130 Hz. Put them all in the oven to dry. An hour later, we re-test B71.2 and find its frequency response un-changed. We do note, however, that reception in the small faraday enclosure with the lid off is 99%, with lid on is 98%, and on our table-top antenna is 100%.

[04-MAR-15] We have three failures in the field that are similar. Two are R64.12 and R64.13. Another is R60.13. The plot below shows an hour when R64.12 is behaving badly a few hours before it expires, and also R64.13 two days before it, too, behaves badly and expires. The behavior of R60.13 was similar.


Figure: Failure of R64.12 and Start of Problems with R64.13.

The rise in the average value of X suggests decreasing battery voltage. The descent in X at the end is a signature feature of the battery voltage dropping from 1.9 V to 1.8 V, as we see here. The battery has been drained prematurely, which suggests excessive current consumption. The fact that R64.13 goes from normal to drained in two days suggests the current consumption is of order 5 mA. Battery drain on its own does not create the step artifacts we see in these failures. The only times we have seen such artifacts is when we have a corroded capacitor. The step artifacts suggest VA is jumping by 0.25 V. Looking at the schematic, if C2 (10 μF), C3 (1 nF), or C4 (10 μF) were cracked, VA would drop by about 0.25V because of the 50-Ω source resistance of the battery. If C5 were cracked, VA would drop to zero, which it does not. If C6 were cracked, VCOM would drop to a fraction of a volt and X would drop down to a fraction of full scale, which it does not. If any of the amplifier capacitors were cracked, we would see huge oscillations on the signal, which we don't, and the battery would not be drained.

The A3028R uses the A302801D circuit board, which places the circuit closer to the center of the battery. As a result, when we clip the programming extension, we have to press the cutters under the board to get to the base of the extension. When we clip, we send a shock wave through the transmitter, which in these batches R60-R71 is already soldered to the battery. Capacitor C4 is next to the cut, and we have been concerned that it would be damaged. But all of R60-R71 have survived three or more days running in water at 60°C, so we thought the possibility of C4 being cracked had been eliminated. Now it appears that this is not the case.

We are changing our assembly procedure. We cut the programming extension off before we load the battery. We load the battery, and the circuit is powered up. We wash in running hot water and scrub, blow dry, and bake. This way, our clipping is less stressful on the board. Our previous concerns about electrical damage when washing a circuit with a loaded battery appear to be unfounded. We were leaving the programming extension connected because it contains a component that disconnects battery power from the circuit. As soon as we clip the extension off, the battery is permanently connected.

[04-MAR-15] We have Test Batch R72.1-10, taken at random from our new lot of 100 A3028R assemblies. Antenna lengths are varied: 30 mm for 1+2, 35 mm for 3+4, 40 mm for 5+6, 50 mm for 7+8 and 60 mm for 9+10. We clip programming extension off before loading battery, wash, blow, bake for 1 hour. All give 100% reception in faraday enclosure, and good mains hum pick-up, except R72.7, which won't turn on.

[05-MAR-15] Transmitter R72.7 draws 2.5 μA from its battery when turned off, and 4 mA when turned on. Its battery voltage drops to 1.2 V. We note that the 0V battery tab is so close to the solder blob of the antenna joint that they may have been in contact in the past. We remove U9, the RF oscillator. On-state current drops to 50 μA. We replace the battery. We wash in running hot water for one minute. We blow dry. We test reception immediately and get 80% in our faraday enclosure. Center frequency is now 922 MHz because of the change in U9. Gain versus frequency is within 1 dB of nominal.

[09-MAR-15] We measure frequency response of test batch R72.1-10. All are within 2 dB of nominal. Reception within the small faraday enclosure is 100% even for transmitters with 30-mm antennas, except for R72.7, which is the single unencapsulated member of the batch, with center frequency 922 MHz. We turn them all on and put them in water to poach at 60°C, except for R72.7, which will bake at 60°C.

[16-MAR-15] We have batch R73.1-14, which has been soaking in water for three days. Frequency response if within 2 dB of nominal for all devices. Reception in faraday enclosure is perfect. Switching noise is less than 4 μV, battery voltage ranges from 2.65-2.73 V. Input noise in water is 8 μV. All turn on and off easily and look good.

We take out Test Batch R72.1-10 after one week running poach at 60°C. All are still running. Battery voltages are around 2.8 V. Frequency response within 2 dB of nominal. Turn on and put back in the oven.

[23-MAR-15] We test batch R72.1-10 after two week running poach at 60°C. All are still running. The figure below shows noise and battery voltages. It appears that VBAT is 2.78-2.82V and noise is less than 8 μV rms.


Figure: Nine Test Transmitters In Water.

Frequency response is within 2 dB of nominal, and appears unchanged from earlier measurements.

[30-MAR-15] We have batch B74.3-12 after four-day inactive soak. Frequency response is within 2 dB of nominal, reception perfect, battery voltage 2.65±0.05 V, switching noise less than 6 μV. When we first start recording, the transmitters are cold and battery voltage is low, so X is clipped at 130 Hz, but once transmitters have warmed up to 20°C there is no more clipping. We re-test B72.1 and B72.2 and they pass all the above tests also.

Test Batch R72.1-10 has endured a three-week running poach at 60°C. Battery voltages are 2.77-2.90 V, total noise is <10 μV in water. Frequency response is within 2 dB of nominal for all except R72.2 and R72.3, which have gain that is 6 dB too low. We drive these with 10 mV through 50 Ω and frequency response is within 2 dB of nominal (R72.2 and R72.3 and compare to R72.5). Reception for all is perfect, magnetic switch is reliable.


Figure: Response of R72.2 with 30 mV, 20 MΩ Source (Left) and 10 mV, 50 Ω source (Right).

It appears that our 10 MΩ input impedance has been reduced by condensation and contamination, so that the 10 MΩ is in parallel with an electrolyte with impedance of order 5 MΩ. Work such as this suggests that the impedance of electrolytes is frequency-dependent, which could explain why the loss of gain at 1 Hz is less significant than from 10-160 Hz. Given that the impedance of most EEG electrodes is less than 100 kΩ, this condensation damage to the circuit will not affect recordings.

APR-15

[03-APR-15] Batch R72.1-10 every transmitter still running after 25 days at 60°C. Battery voltages are 2.78-2.93 V and noise is less than 8 μV.

[06-APR-15] Batch R72.1-10 are still running. We get no reception from R72.3. Center frequency is 927 MHz. Encapsulation is in perfect condition. We remove silicone and scrape epoxy off the battery terminals. Battery voltage is 2.8 V. Active current is 82 μA, inactive is 2.7 μA. We measure center frequency of the remaining transmitters. All are within 915-919 MHz, except R72.7, which remains 920 MHz as before. We measure frequency response of all but R72.3 and all are within 2 dB of nominal, except R72.2, which still has −6 dB gain when driven by 20 MΩ source.


Figure: Spectrum of R72.3 After Failure of Reception.

The spectrum of R72.3 has the correct width and power. We can see traces of side lobes on either side of the center. It was programmed originally with f_low = 8. The side lobes are at 927±4.5 MHz, which is the modulation width we expect with ADC count changing by 2, in this case from 8 to 10. Each count gives us between 4-5 MHz increase in output frequency. The entire spectrum has moved up by 8 MHz from its original calibration, which is like adding 4 to the original 8, which is consistent with DAC output bit 2 being stuck HI when it should be LO.

An hour later, the peak frequency drops suddenly to 922 MHz and we can get some reception from the device. We put it back in the oven at 60°C in a bag.

We have circuit board B9883.0014. We program it with P3028A05 as an A3028C with f_low=8 and it works fine, with f_low=7 we have the F2 output behaving incorrectly during transmission. This is on our calibration PC with ISP Lever 1.4. We move to a lap-top running ISP LEver 1.7 to study the problem some more, but the problem does not arise. We go back and forth between machines and the fault appears with V1.4 and not with V1.7.

[10-APR-15] The compiler version turned out to be irrelevant to the timing problem. We modified the way we did the frequency control, introducing the FHI node in P3028A06 and now the DAC output is correct. But when we disable the test points, which we do out of curiosity, the current consumption of the chip jumps up to 50 mA. We find that when we set the test point outputs to zero, this combines with certain combination of ID, fck_divider, and frequency_low to produce a 50-mA current drain. We cannot understand why this happens. Our best guess is a compiler bug in the Lattice 1.8 compiler. We remove any mention of TP2 and TP3 from the code and set TP1 to FHI. We program ten circuits with different parameters and all seems well.

We take our R72 test batch out of the oven, cool them down, and find all transmitters are running. We do not check battery voltage or frequency response. We put them back in the oven again.

[13-APR-15] We take out Test Batch R72.1-10. R72.4 and R72.9 are not transmitting. We see no response from the Data Receiver or the Spectrometer. All others are transmitting and being received. We remove silicone from R72.9. No sign of breech in encapsulation. Battery voltage is 0.7 V, current 200 μA. We disconnect battery and voltage rises to 2.0 V. With 2.7 V power supply the magnetic switch still works, with current at 2.0 μA when off. When on, current varies 3-4 mA over the a fraction of a second. We remove silicone from R72.4. No sign of breech. Battery 0.7 V, quiescent current 200 μA. Disconnect rises to 1.4 V. With 2.7 V off current is 2.0 μA, on current varies from 5-50 mA over a fraction of a second.

The variation in the quiescent current is a symptom of a corroded capacitor. In both cases, the off-current is correct, and the on-current exceeds 4 mA. Looking at the schematic, We see that the damaged capacitor cannot be C2, because it would always be draining the battery. It cannot be C5 or C6 because they have 1 kΩ resistors in series, which limit their current to 2.7 mA and 1.8 mA respectively. It cannot be any capacitor in the amplifier, because the amplifier runs off VA, which has a 1-kΩ series resistor. It must be either C3 or C4. We remove C3 and C4 from R72.9 and on-current is 3-4 mA. We remove C3 and C4 from R72.4 and on-current is 2-20 mA.

We remove U9 from R72.9 and current is 2-3 mA, and from R72.4 and current is 150 μA for a few seconds, then jumps up to 2 mA and stays there. We turn it on and off a few times and observe the same pattern, although one time it starts up with quiescent current around 50 μA. We remove C5 from R72.4, 3 mA. We remove R3 and R4 from R72.4, 300-400 μA, but we wait only ten seconds. We remove R3, R4 from R72.9, 3-4 mA.


Figure: R72.4 and R72.9 After Part Removals. Current consumption is still excessive in both circuits.

The only chips left with power are U2+U3, which are working fine, and U4+U8+U10. One of these parts is responsible for erratic and excessive current consumption on both R72.4 and R72.9. We recall the varying current consumption of the logic chip when over-heated during construction here.

R72.1, R72.5, R72.6, R72.8, and R72.10 have gain versus frequency within 2 dB of nominal. Battery voltage is 2.75-2.85 V, noise is 7 μV. R72.2 transmits only zeros. We remove silicone. Battery voltage is 2.6 V. Current consumption when off is 40 μA and when on is 2.2 mA. R72.3 is running well, with center frequency 919 MHz, but it is no longer encapsulated. R72.7 is running well, with center frequency 925 MHz as before, but it is no longer encapsulated.

We have batch C74.13-76.1, 256 SPS, 80-Hz bandwidth single-channel mouse transmitters. We measure and record frequency response to 8-s, 1-300 Hz sweep. All look good except R75.10, which behaves as if one of the electrode leads is not connected at the circuit board. We place in hot water to soak.

[14-APR-15] We find several discussions of dendrite formation, such as A Review of Models for Time-to-Failure Due to Metallic Migration Mechanisms. These papers suggest that the rate at which metallic dendrites form on the surface of a circuit board, causing short circuits, is a strong function of temperature. In the Hornung model, the mean time to failure once a water film has formed, is given by:


Where T is absolute temperature, s is the spacing of the pads to be joined by a dendrite, E is the activation energy of the dendrite-forming reaction, once a water film has been formed to permit the reaction, in units of eV, and k is the Boltzmann constant, 8.6 × 10−5 eV. Hornung found that E ≈ 1.1 eV for silver on glass. We assume a similar value for dendrites between the pads of U4, U8, or U10.

The minimum track spacing between on our A302801E circuit board is 5 mils, or 125 μm. Dendrites are more likely to form between exposed pads than between tracks covered by solder mask. We believe the current drain must be taking place in and around U4, U8, or U10. The spacing between the bads under U4 is 200 μm and there are 2 gaps, of which only one will cause problems if shorted. Under U8 the spacing is 175 μm and there are 56 such gaps, of which 16 will cause problems if shorted. Beneath U10 the gap is 500 μm. According to the Hornung model, dendrites are ten times more likely to form under U8 than under U4 and U10 combined.

In our experience, it takes a few days at room temperature for water films to establish themselves within our encapsulation. The water films will exist only where our epoxy coating is not bound to the solder mask and pads. The most likely place for openings to exist between pads is beneath U8, which is a 0.5-mm pitch ball grid array. It is beneath this part that air has the greatest distance to travel during the evacuation phase of our encapsulation. Our R72 test batch has been in water at 60°C for five weeks, with half a dozen cool-downs during that time, which will provoke condensation. Let us assume that water films formed early on, so it took most of five weeks for dendrites to grow between the pads beneath U8.

On the subject of activation energy, this page says, "Silver is the metal most susceptible to migration, since it is anodically very soluble and requires a low activation energy to initiate the migration process. Copper, zinc, and lead will also migrate, although only under much more severe conditions. Most other common electronic materials are not susceptible to migration: iron, nickel, and tin because of their low solubility in water; gold, platinum, and palladium because they are anodically stable." These circuit boards were made with lead-free solder, which means they contain silver. Thus the activation energy of dendrite formation will be of order 1.1 eV.

According to the Hornung model, if E = 1.1 eV and the time to failure is 5 weeks at 60°C, the time to failure at 37°C, which is the approximate temperature in a rat, will be 14 times longer, or 70 weeks. At 20°C, the time to failure will be 100 times longer, or 500 weeks.

[15-APR-15] Test transmitters R72.1, 3, 5-8, and 10 are still running. Battery voltages all normal.

We take out R72.4 and pry the circuit board off the battery. The black encapsulating epoxy holds on to the parts, so that the component footprints come off the circuit board. The picture below shows the balls of the BGA surrounded by epoxy, with traces of solder mask that have been pulled off as well.


Figure: Epoxy Surrounding BGA Balls After Prying Battery from Circuit Board. We see copper pads with ENIG plating torn from the circuit board by the epoxy, taking the copper tracks with them.

We examine the remains of the footprint on the printed circuit board.


Figure: BGA Footprint After Prying Battery from Circuit Board. We see the pads torn away by the solder balls.

We are delighted by the penetration of the epoxy under the BGA. We see no sign of dendrites. According to this treatise, integrated circuits in plastic packages can suffer damage in high humidity and heat whether they are biased (powered up) or unbiased (no power connected). The biased circuit moisture test performed by Lattice Semiconductor, manufacturer of U8, is to subject the package to 85°C and 85% relative humidity for 1000 hr. In Reliability Technology: Principles and Practice of Failure Prevention in Electronic Systems by Pascoe, the author presents Hallberg and Peck's formula for rate of failure in integrated circuits. The rate is proportional to the third power of relative humidity, and to exp(E/kT) with E = 0.9 eV. Thus 25 years at 20°C and 50% humidity is like 32 days at 60°C and 100% humidity or 6 days at 90°C and 80% humidity. The third conditions caused failures similar to the first condition. Given that U8 can survive 1000 hr at 85% humidity and 85°C, it can endure 4500 hr at 60°C and 100% humidity.

We note, however, that Hallberg and Peck's work does not cover the case where the components are immersed in water. Our test batch is poaching in 100% humidity at 60°C with two quenches in cold water per week, which will cause condensation in all available cavities within the encapsulation. When immersed in water, printed circuit board traces can be joined by dendrites in a few minutes at 85°C. (This video appears to show rapid dendrite formation in hot saline, although we are not certain that it is real-time.)

If we cannot use the absolute formula for time to failure that is provided by Hallberg and Peck, we can use their study of the activation energy of various rate-determining steps in the chemical process that leads to failure. They give this energy as close to 0.9 eV. We refine the calculation we did above, using 310 K for implanted devices and 333 K for our test batch. We get a time to failure 10.3 times longer for implanted than poaching at 60°C. Our first failures occurred after 5 weeks at 60°C, so we expect time to failure while implanted to be closer to 50 weeks.

[16-APR-15] Batch C74.13-C75.9, C75.11-C76.1 has been soaking in water since 13-APR-15. Switching noise is less than 4 μV for all devices and VBAT is 2.55-2.65 V. When measuring frequency response in the Neuroarchiver, we set the default frequency to 256 SPS this time. Frequency response is within 2 dB of nominal for all devices.

[17-APR-15] Test transmitters R72.1, 3, 5-8, and 10 are still running. Battery voltages all normal.

[22-APR-15] We have batch R76, which has been soaking in hot and cold water for four days. Battery voltage 2.58-2.68 V in cold water. Switching noise less than 12 μV. Switching noise less than 5 μV. Frequency response within 1 dB of nominal except R76.8, which is 2 dB above nominal at 130 Hz.

Test transmitters R72.1, 3, 5-8, and 10 are still running. Battery voltages are normal except R72.6, which has dropped to 2.38 V. Gain versus frequency in R72.1 is 3 dB too low at 50 Hz and 6 dB too low at 130 Hz. In R72.5 it is 3 dB too low at all frequencies. The others have gain versus frequency within 2 dB of nominal.

[27-APR-15] Test batch R72.1-10 were put in to poach at 60°C 7 weeks ago today. We are ending the test today, after the equivalent of 70 weeks at 37°C. R72.1 developed reduced input impedance at Week 6. At Week 7 it functions perfectly when driven by a 50-Ω source. Battery voltage 2.8 V. R72.2 developed reduced reduced input impedance at Week 3 and failed by all-zero transmission then battery drain at Week 5. R72.3 develops reduced input impedance at Week 3 and a +10 MHz shift in center frequency at Week 4. This shift reversed after we tore off silicone. We baked until Week 7 and it functions perfectly with battery voltage 3.0 V. R72.4 stopped transmitting at Week 5 and dissection showed excessive current drain. R72.5 developed reduced input impedance at Week 6 and battery voltage has dropped to 2.2 V at Week 7. R72.6 battery drain took place in the days leading up to Week 7. R72.7 we baked the entire seven weeks and is running perfectly still, except for the poor frequency calibration it started with. R72.8 drained its battery some time during Week 7. R72.10 developed reduced input impedance at Week 7.

The reduced input impedance would not affect recordings from 10 kΩ electrode wires or 1 kΩ skull screws. The first significant failure we observe is the temporary frequency shift in R72.3 at Week 4. Battery drain problems start at Week 5. At 37°C we expect our time to first failure in a batch of 10 transmitters to be around 40 weeks, which is longer than the 18-week operating life of the A3028R.

[06-MAY-15] We have batch R77.10-78.9 after a seven-day inactive soak in hot and cold water. Gain versus frequency is within 1.5 dB of nominal for all. Reception is perfect. Battery voltage and noise are normal.

MAY-15

[08-MAY-15] We have batch R76.14-77.9 after a four-day soak with charges of hot water. We turn them all on and place them in water in a faraday enclosure with two pick-up antennas.


Figure: Batch R76/77 Reception, Battery Voltage, and Noise. Transmitters all in water, running at the same time.

The imperfect reception is due to collisions in the small space of the enclosure. When we place the transmitters individually in the enclosure, reception is in every case perfect. We measure gain versus frequency and find it to be within 1 dB of nominal with the correct slump in gain before the bump at 130 Hz.

[11-MAY-15] We put a plastic bin in our FE2A faraday enclosure. We tape a transmitter to a self-propelled ball. Later, we put the ball in a latex glove with the transmitter and allow it to move around in water, like this. We measure reception from one antenna with the lid on and off, for various transmitters, over a one-minute period.

DeviceLidAntennaReceptionImmersedLeads
B79.1on30 mm98.3no100 mm, 1 loose, 1 taped
B79.1on30 mm97.1no100 mm, 2 taped
B79.1on30 mm97.9no100 mm, 2 loose
B79.1off30 mm73.3no100 mm, 2 taped
C75.10on50 mm99.1no45 mm, 2 taped
C75.10on50 mm99.0no45 mm, 2 loose
C79.2on30 mm98.8no45 mm, 2 loose
R76.13on50 mm99.1no150 mm, 2 taped
R76.13on50 mm99.1yes150 mm, 2 taped
C75.10on50 mm94.2yes45 mm, 2 taped
B79.1on30 mm97.7yes100 mm, 2 taped
B79.2on30 mm93.4yes100 mm, 2 taped
B79.2on30 mm95.3yes100 mm, 2 taped
Table: Reception for Various Antenna Lengths while Moving in Air and Water.

We measure received power from moving transmitters also. We record the average decibel power measurement (dBm is 10 times log of power divided by 1 mW), min and max power also.

Device Antenna Min
Power
Ave
Power
Max
Power
Immersed
B79.2 30 mm −61 dBm −47 dBm −36 dBm no
B79.2 30 mm −68 dBm −49 dBm −34 dBm no
C75.10 50 mm −66 dBm −47 dBm −35 dBm no
R76.13 50 mm −68 dBm −48 dBm −37 dBm no
R76.13 50 mm −75 dBm −54 dBm −39 dBm no
C75.10 50 mm −73 dBm −57 dBm −39 dBm no
B79.1 30 mm −75 dBm −54 dBm −38 dBm no
Table: Power Received for Various Antenna Lengths while Moving in Air and Water. The lid of the faraday enclosure is always on.

We see no significant difference in performance between immersed transmitters with the 30-mm and 50-mm antennas. This suggests that we might reduce the length of the mouse transmitter antenna without compromising performance. We see no difference between power received from rat and mouse transmitters, so we still have no explanation for why mouse transmitters perform less well in IVC racks than rat transmitters.

[11-MAY-15] We have test batch B78.10-B79.5. They are all encapsulated with 30-mm antennas. Device B78.13 is running when we receive them for first test. Center frequency is 914-919 MHz. Reception 100% in faraday enclosure for all. Gain versus frequency within 2 dB of nominal. We place in the oven at 60°C in a box for burn-in in dry air.

[12-MAY-15] Eight out of ten transmitters in text batch B78.10-B79.5 are running when we drop them in a pile on our bench antenna. Two are not running. We turn them on. They are B78.10 and B79.4. We place them all in water. At first, battery voltages vary, but after ten minutes we get the following with one antenna in a faraday enclosure.


Figure: Batch B78/79 Reception, Battery Voltage, and Noise. Transmitters all in water, running at the same time.

We take B78.14 out to bring to ION for some tests. We leave the other nine running in water at 60°C.

[18-MAY-15] Batch B78.10-13, B79.1-B79.5 all running fine after one week running poach. We check frequency response, reception, and battery voltage.

[20-MAY-15] Batch B78.10-13, B79.1-B79.5 all running. B79.1 battery voltage is 2.31 V. Others are 2.62-2.69 V. Frequency response of all devices is okay. Center frequencies 915-920 MHz. B79.1 we turn on and off a few times. Its battery voltage remains the same. Return to oven to continue running poach.

[21-MAY-15] Batch B80.1-11, B80.13-14 has passed through 24hr/60°C/D/ON (twenty-four hours at sixty degrees centigrade in dry air, running). Battery voltages are 2.49-2.53 V. Noise is 10 μV. Reception is good. Gain versus frequency close to nominal. We place in water for 72hr/20°C/W/OFF.

[21-MAY-15] Batch B78.10-13, B79.1-B79.5 has been running in water at 60°C for ten days. Two won't turn on: B79.1 and B79.2. The others are still running and battery voltages are normal. We remove silicone from B79.1, cut through positive battery lead, scorch away epoxy around negative battery lead, and connect 2.7 V. The device switches on and off, transmits well, and responds to mains hum. The average value of X varies correctly with VB. Current consumption while running is 370 μV, and when off is 300 μA. Remove C2, active current consumption drops to 80 μA. We expose B79.2 battery terminals and disconnect battery. Its voltage is 2.1 V. Connect 2.7 V. When on, transmitter draws 90-120 μA, varying. When off, 2.1 μA. We turn on and wait ten minutes. Current is stable at 89 μA. No sign of corrosion, leaking, or discoloration of any part of the encapsulation or the leads. We note that scorching off the epoxy and soldering test leads to the battery footprint heats up the entire circuit.

[22-MAY-15] B78.10 has stopped. B79.5 battery voltage 2.14 V. B78.11-13,B79.3-4 have battery 2.54-2.63 V. They pick up mains hum. We expose B79.5's battery terminals. We turn off and measure VB−VC = 2.4 V. We cut battery positive lead and measure VB 2.5 V. We apply 2.6 V. Off current 1.6 μA, on current 81 μA. Expose B78.10 terminals. Cut positive battery lead, measure VB−VC = 1.4 V. Apply 2.6 V. Off current 1.7 μA. On current 1.7 mA. Transmits zeros. We heat up C5 with a soldering iron. On current drops to 90 μA. We raise VB to 4.3 V. On current 104 μA. Drop to 2.6 V, 89 μA. Average X now indicates VB = 2.88 V. No sign of corrosion, leaking, or discoloration of any part of the encapsulation or the leads.

We take a fresh A3028 circuit and calibrate it. We replace C2, C5, and C6 with 100 nF. We solder the X inputs together and measure the following noise spectrum over 16 s.


Figure: Noise with C2 = C5 = C6 = 100nF and Inputs Shorted. Noise amplitude is 9.4 μV rms.

So far as we can tell, dropping the decoupling capacitance on VB, VA, and VC from 10 μF to 0.1 μF has no affect on the signal noise. Gain versus frequency is within 2 dB of nominal after we wash and dry.

[24-MAY-15] B78.11, B78.12, B79.3 still running with battery voltages 2.58-2.64 V. But B78.13 and B79.4 have stopped. B78.13's battery, when disconnected, is 0.8 V. Inactive current with 2.6 V applied is 1.6 μA and active is 83 μA. Picks up mains hum. Reception is good. B79.4's battery, when disconnected, is 1.7 V. Inactive current when we first apply 2.6 V is 3000 μA, but we disconnect and re-connect and now active current is 89 μA and inactive current is 1.9 μA.

[26-MAY-15] B78.11 battery voltage is now 1.98 V while B78.12 is 2.67 V and B79.3 is 2.65 V. All pick up mains hum. All switch on and off. We strip silicone off B78.11 and measure VB−VC = 0.3 V. We disconnect the battery and apply 2.6 V. Active current 92 μA, inactive 2.1 μA.

[27-MAY-15] Batch B80.1-11, B80.13-14 has been through 120hr/20°C/W/OFF. Transmitters B80.5 and B80.11 have rust around their positive battery terminals, as shown below. We observe the rust re-forming twice, so we are convinced it is not debris in the water.


Figure: Rust Around Positive Battery Terminal After 120hr/20°C/W/OFF.

We observe water leaking out near the positive battery terminal in B80.11. There is one bubble near the positive battery terminal on each of B80.5 and B80.11. The silicone over the corners of the battery terminal is thin in B80.1. B80.5 switching noise is 80 μV, B80.11 switching noise is 40 μV, B80.1 switching noise is 24-μV switching noise and harmonics. Remainder have switching noise less than 3 μV. Battery voltages are 2.53-2.56V. Frequency response of all devices is within 2 dB of nominal. We connect a microammeter to C and a pan of water. We lower the body of the device into water and measure current flowing from the battery positive terminal to V through whatever leak might exist in the silicone coating. The transmitter is off. If there is current, we confirm that it is the corner with the battery terminal that is the source of the connection to the water. We test all devices. B80.1 = 2 μA, B80.5 = 20 μA, B80.11 = 15 μA, all others 0.0 μA.

B78.12 and B79.3 are still running with normal battery voltage.

We have batch B79.6-9, B79.12-B80.4. These were intended to be a batch of A3028Rs, but we left them soaking in water for five days and we now find there is white residue around the logic chip. We decide to replace the 10-μF capacitors with 1-μF capacitors and load mouse batteries to see if we can get longer life with the smaller capacitors.

[28-MAY-15] Batch B80.1-11, B80.13-14 now has two extra coats of silicone over the positive battery terminals. We measure battery leakage through encapsulation and find it 0.0 μA for all devices. Noise in B80.1 is 9 μV rms, in B80.5 is 12 μ, and in B80.11 is 8 μV.

B78.12 still running with normal battery voltage, but B79.3 has died. Its battery voltage is 2.4 V. Inactive current is 1.7 μA. Active current is 4 mA and we get reception. We remove C4. Active current is 3 mA and we have no reception. We re-connect power and active current is 100 μA but no reception. We replace C4. Active current 93 μA and good reception.

[29-MAY-15] B78.12 still running with VB = 2.7 V. Noise is 24 μV. Switching noise is 8 μV.

JUN-15

[01-JUN-15] B78.12 still running with VB = 2.65 V. Noise is 14 μV. Reception and frequency response are normal. Today is its 19th day poaching.

[03-JUN-15] B78.12 still running with VB = 2.43 V. Noise is 16 μV rms.

[04-JUN-15] B78.12 has expired after 23 days running at 60°C in water. We write this e-mail summarizing our reliability studies.

[05-JUN-15] We have B79.6-9,B79.12 with 1-μF capacitors in place of C2, C4, C5, C6, and C8. batteries loaded. We have B79.13-B80.2 with the 10-μF capacitors intact. These are protected-input circuits on the A302801D circuit board. We load 48 mA-hr batteries. Three of them, B79.9, B79.12, and B80.1 won't turn on, although they worked when powered earlier in the day from the programming extension. We observe dendrite growth between U1-1 and U1-2, which turns off the U1-1 mostfet so VM is disconnected from VB. No power reaches the circuit.


Figure: Dendrites Between U1-1 and U1-2 After Five-Day Room-Temperature Soak. The dendrites are the ivy-like connection between the rightmost pads.

We scrub the boards and dry in hot air. All three of them now work well, picking up mains hum and giving good reception.


Figure: Pins U1-1 and U1-2 on B80.1 After Cleaning.

We inspect the remaining circuits, and they too have corrosion and dendrite growth around U1-1 and U1-2. But there is no connection between the pads yet.

[18-JUN-15] We hear of the sudden failure of R69.3 at Marburg, in archive M1434591227.ndf. Reception starts to degrade at 3:39:47. By 3:39:50 reception has stopped. We are back to 100% reception at 3:40:03. From there, reception is close to 100%, with stable battery voltage until 3:49:51, when reception drops to 0% from one sample to the next, with no change in battery voltage in the final moments.

[23-JUN-15] Test batch No79 consists of transmitters B79.6-9,12-14, B80.1-2. Transmitters B79.6-9, 12 we equipped with 1-μF in place of C2, C4, C5, C6, and C8. The remainder are control circuits with 10 μF in these locations. The circuit boards are A302801D input-protected circuits with mouse batteries loaded, to make something equivalent to the A3028B, but looking as below.


Figure: The A3028B Made with Input Protection Circuit Board.

The protruding battery battery contacts will allow us to gain access to the battery terminals without having to heat up the capacitors on the circuit. B79.7 fails after encapsulation. The remaining eight survive burn-in 24hr/60°C/D/ON. We measure noise in water less than 12 μV and reception is perfect. Center frequencies are around 915 MHz. Frequency response within 2 dB of nominal except No79.6 and No79.8 which show +3 dB at 130 Hz. Note that in B79.6, B79.8, B79.9, and B79.12 we have C8=1μF and R6=50kΩ, which gives the transmitters a high-pass response with half-power frequency 3 Hz. The other four have low-frequency cut-off at 0.3 Hz. The difference in their response is clear in the first few oscillations of our 1-500 Hz sweep. We turn them all on and put them in the oven to poach in water at 60°C.

[24-JUN-15] ION measures reception in an IVC rack isolation chamber. They arrange four antennas around four mouse cages, each with a mouse and A3028B transmitter. Two other cages are on the far edges of the rack. They measure reception during four hours with an Antenna Combiner (A3021B) and Data Receiver (A3018D) and then for four hours with an Octal Data Receiver (A3027D).

NoA3021B+A3018DA3027DComment
494.797.0Near
678.198.3Far
996.698.8Near
1297.796.7Near
1397.499.4Near
1480.298.2Far
Table: Reception in New IVC Rack.

We still have not figured out why mouse transmitters performed less well than rat transmitters in ION's original IVC rack enclosure, when a 20-cm gap was present in the back corner. But we see that reception from these transmitters is excellent in the new enclosure without gaps.

[25-JUN-15] Transmitter B82.4 has a 30-mm antenna. We place it on out spectrometer antenna and measure peak power −29 dBm at 915 MHz. We place it in a petri dish of water in the same location and get −39 dBm.

Batch B82.4-14 have endured 24hr/60°C/D/ON. Frequency response is within 2 dB of nominal for all. Battery voltages 2.51-2.66 V. Noise below 15 μV. Reception robust. Turn them all off and put them in water at room temperature.

Test batch No79 all eight running, battery voltages 2.64-2.73 V. Reception from jar is good after cooling down. Return to oven to continue poach.

[29-JUN-15] Batch B82.4-14 has endured 96hr/20°C/W/OFF. Frequency response within 2 dB of nominal, battery voltages 2.48-2.62 V, noise below 15 μV, reception robust. Turn them all off and dry them ready to ship.

Test batch No79 all eight running. Battery voltages 2.67-2.75 V. Noise less than 15 μV except B79.8, which has noise 35 μV in repeated measurements. The noise spectrum of all eight transmitters is here. B79.8's noise appears to be switching noise plus some higher-frequency noise similar to what we see when the 0-V wire breaks. We measure frequency response. All within 2 dB of nominal except B79.8 has gain 3 dB above nominal at 130 Hz. Reception is robust for all. All are running when we return them to 60°C poach.

JUL-15

[01-JUL-15] Test Batch No79, eight transmitters still running, battery voltages 2.64-2.74 V.

[02-JUL-15] Test Batch No79, eight transmitters still running, battery voltages 2.67-2.74 V.

[03-JUL-15] Test Batch No79, eight transmitters still running, battery voltages 2.63-2.68 V.

[04-JUL-15] Test Batch No79, eight transmitters still running. B79.8 has 20-mV 195-Hz oscillation that we cannot stop even by shorting X leads together, reception perfect. B79.2, B79.6, B79.9, B79.12, B79.13, B79.14 gain versus frequency within 2 dB of nominal, reception perfect. B80.1 gain is 3 dB too high at all frequencies, reception perfect. Battery voltages 2.64-2.72 V for all. Noise less than 20 μV in all but B79.8. Turn them all on and continue the poach.

[06-JUL-15] Test Batch No79, eight transmitters still running. Battery voltages 2.64-2.72 V after a few minutes settling at 20°C. During settling, No79.13 shows jumps of order 1 kcount and No80.2 of order 100 counts.

[08-JUL-15] Test Batch No79, eight transmitter still running. Average value of B79.12 starts off at 51 kcount and drops over twenty minutes to 46 kcount. During this time, the oscillations on B79.8 diminish and vanish over a one-minute period. We now get the following.

ID (No) RECEPTION (%) VA (V) NOISE (uV rms)
1 98.83 2.66 9.96
2 94.34 2.71 40.36
6 98.83 2.64 18.48
8 96.29 2.53 35.36
9 90.62 2.64 18.20
12 99.80 2.56 18.68
13 99.41 2.69 7.96
14 80.66 2.67 9.52

Batch R83.2-12 endured 24hr/60°C/W/ON. All are fine afterwards except R83.4, which won't turn on or off. Turn them all off and put them in water to soak.

[10-JUL-15] Test Batch No79, eight transmitters still running. We put them in 20°C water. B79.6 noise 15 μV, VBAT = 2.6 V. B79.8 oscillating initially, but settles down to 150 μV noise and VBAT = 2.4 V. B79.9 noise 25 μV, VBAT = 2.3 V. B79.12 oscillating full-scale at 160 Hz. B79.13 occasional 10-mV swings. Otherwise battery voltage appears to be 2.6 V. B79.14 shows occasional 10-mV swings. Otherwise battery voltage appears to be 2.6 V. B80.1 shows occasional 10-mV swings. Otherwise, battery voltage appears to be 2.6 V. B80.2 shows occasional 10-mV swings about a baseline. Battery voltage appears to be 2.1 V.

[11-JUL-15] Test Batch No79, seven transmitters still running, B80.2 won't turn on. Battery voltages and noise are as shown below. B79.12 is still oscillating at 160 Hz.

ID (No) RECEPTION (%) VA (V) NOISE (uV rms)
1 97.46 2.64 12.00
6 97.46 2.64 19.88
8 99.80 2.20 257.56
9 99.02 2.61 19.12
12 100.00 2.67 5619.84
13 94.92 2.70 16.84
14 94.92 2.63 17.76

We measure current consumption of a bare A3028A circuit with input leads shorted together and get 90 μA. We connect 100 mVpp, 100 Hz and get 95 μA. With 100 mVpp, 10 Hz, 94 μA. With 1 Vpp, 100 Hz we get 94 μA. Back to inputs shorted we get 89 μA.

[13-JUL-15] We have batch R83.2-12. Of these, R83.4 already failures during burn-in. After 6days/20°C/W/OFF the ten remaining have gain versus frequency within 2 dB of nominal. Noise and reception simultaneously in water are as follows.

ID (No) RECEPTION (%) VA (V) NOISE (uV rms)
2 48.83 2.66 11.08
3 50.00 2.67 8.36
5 48.63 2.63 11.76
6 48.05 2.70 14.40
7 48.63 2.67 12.44
8 48.05 2.68 16.24
9 50.00 2.70 10.96
10 45.51 2.68 10.32
11 49.80 2.65 15.08
12 48.24 2.67 13.56

Test Batch No79, seven transmitters still running. R79.6 gain versus frequency has nominal shape but is 2 dB too low throughout. R79.8 still oscillating at 160 Hz. R79.9 baseline swings and noise. Gain has correct shape but is 6 dB too low. R79.12 gain is 10 dB too low and when in water we see spiky noise of 7-15 Hz, usually around 10 Hz, as shown below.


Figure: In-Water Self-Generated Noise of R79.12. The device has been running in water at 60°C for twenty days.

R79.13 gain is 6 dB too low. R79.14 and R80.1 gain is within 2 dB of nominal. Battery voltages are 2.46-2.70 V.

[15-JUL-15] Test Batch No79, two transmitters still running. B79.8, B79.9, B79.12, B79.14, and B80.1 won't turn on. B79.6 battery voltage now around 2.3 V but it picks up mains hum just fine. B79.13 still has lots of low-frequency noise and rumble. Battery voltage appears to be around 2.7 V. We turn off the transmitter, remove silicone and measure battery voltage directly as 2.7 V. We put B79.13 in the oven. After an hour, noise is gone, gain is 6 dB too low but otherwise looks good.

Test Batch No79 Summary: These eight transmitters were built with the A3028R1 protected-input circuit, built with lead-free solder in January 2015, and loaded with 48 mA-hr batteries. Expected operating life is 21 days. All eight suffered corrosion of joints when left in water for several days. We expected problems with the EEG amplifier as a result, but our objective was to look for failure in the 1-μF and 10-μF capacitors. Four had 1-μF capacitors in place of 10-μF. Of these four, all ran for at least 21 days. One developed oscillations in its amplifier and another showed loss of gain. Of the four with 10-μF capacitors, one failed after 18 days. The others ran for over 21 days.

[16-JUL-15] We are having trouble with the P3028A05 firmware. Once again, the problem arises with frequency_low = 7, as we observed with firmare 4 and 5, and which we believed we had fixed in 6. We note that the TCK period is not increasing linearly with fast clock divisor, and from 13 to 14 it actually decreases. In our P3030D07.vhdl firmware, we discovered that the failure of the fast clock divisor was the result of the ring oscillator being too fast. We add another gate to the ring oscillator, so it now has three gates and a frequency of around 100 MHz. At this lower speed, the clock divider is stable, and we obtain the following encouraging plot of TCK period versus fck_divisor.


Figure: TCK Period versus Fast Clock Divisor. We show A3028AV4 hardware with firmware V2 and V7. The period should lie in the range 195-220 ns for perfect reception. We later add plots for newer circuits.

The transmit clock (TCK) generation is well-behaved, but the problem with frequency_low = 7 persists.

[29-JUL-15] In the A06 firmware, we declare signal FHI as a pin with no assigned location. This is a bug we introduced in A06 as we were trying to fix another bug. If we make FHI a node that we keep, the problem with frequency_low stops. But we still have glitches on FHI so we replace it with BIT, which we keep as a node so as to simplify the calculation of F1..F4. If we don't keep BIT then F3 comes out wrong for frequency_low = 7 and x_id = 7 with version = 1. We re-program 14 A3028B circuits with this new code, P3028A07. The average current consumption without antenna drops from 85.8 μA to 78.9 μA, which for the A3028B means expected battery life rises from 559 hours to 608 hours.

[31-JUL-15] We take an A3028D and replace all its 10-μF capacitors with 2.2 μF. We set R6 = R13 = 249 kΩ, R7 = R14 = 10 MΩ, and C9 = C14 = 220 pF. We measure gain versus frequency for both X and Y.


Figure: Amplitude versus Frequency for 10-mV Input, 2.2 μF Capacitor Circuit.

We increase R6 and R13 from 50 kΩ to 249 kΩ so as to keep the time constants R6C9 and R13C14 at 0.5 s. We increase R7 and R14 from 2 MΩ to 10 MΩ so as to keep the gain of the first amplifier stages at ×40. Ideally, we would drop C9 and C14 from 1 nF to 200 pF to keep the time constants R7C9 and R14C14 at 10 ms. But we don't have 200 pF so we use 220 pF instead. One concern we have about moving from 10 μF down to 2.2 μF is noise on the inputs from poorly-decoupled power supplies. When placed in a faraday enclosure with inputs connected by 20 MΩ, we see 36 counts of noise, or 14 μV.

We have batch B84.1-14, all with 30-mm antennas, our new standard. After epoxy, B84.9 will not turn on, but battery voltage is 3.0V. After encapsulation, frequency response of the remaining circuits is within 2 dB of nominal. Reception is 100%.

AUG-15

[10-AUG-15] Batch B84 has endured 2hr/60°C/D/ON and 72hr/20°C/W/OFF. Frequency response is within 2 dB of nominal, reception is 99% or higher, battery voltages 2.51-2.51V, noise around 12 μV.

We inspect a B85.9, which failed after loading the battery, and find that the positive battery terminal is not soldered. We solder the terminal and the transmitter now turns on and off, has good reception and nominal frequency response. We inspect B84.9, which failed after encapsulation. We remove epoxy from the positive battery terminal and find it is not soldered. We solder it and the transmitter now turns on and off, has good reception, and nominal frequency response. We have one bare A3028R1 assembly that won't turn on and one A3028AV3 assembly that won't turn on. We find that U3 is not responding. We cannot remove U3 by heating, so we tear it off another board without heating first, so as to make sure the underneath is clean. We get this photograph of the underside, in which the PCB tracks have adhered to five of the pads, but not to U3-2, which will give rise to failure to turn on and off. Beneath the one we heated, we have a sticky mess, which might be dried out flux.

[12-AUG-15] We hear from our assembly house that U3 would be better soldered with their My600 paste printer. We also resolve to ask for leaded assembly rather than unleaded.

We have batch B85.1-14. Of these, B85.9 failed because its battery tab was not soldered, B85.14 failed for some unknown reason because we can't find it to test it, and the remainder have endured 2hr/60°C/D/ON and 72hr/20°C/W/OFF. Frequency response is within 2 dB of nominal. Reception is 99% or higher. Battery voltage 2.52-2.61 V, noise around 10 μV in water.

SEP-15

[04-SEP-15] We have 40 of the new A3028AV4, following the A3028D_1 schematic, with 2.2-μF automotive-grade capacitors (in place of 10-μF general-purpose) and leaded solder (instead of lead-free, as we called for in the A3028AV1-3). We measure gain versus frequency for X input and find it within 1 dB of nominal. We obtain this plot of period versus divisor. The ring oscillator is running at around 73 MHz, compared to 92 MHz for the V3 hardware.

[08-SEP-15] We are working on a depth electrode for Iris Oren, which we will call Electrode H. The idea is to solder the EEG lead directly to a teflon-insulated, platinum-indium electrode wire, at a 90° angle. The implanter will cut the Pt-In wire to the desired length. A holder attached to the electrode allows us to raise and lower it until it is located correctly. We cement the electrode in place, then remove the holder. The figure below shows our prototype, in which we have soldered three wires together, one Pt-In 125-μm diameter insulated with 200-μm diameter teflon, one stainless steel helix in red silicone, and one 260-μm diameter tinned copper wire in 530-μm PVC. The copper wire runs up through a blunt-ended syringe needle. When it emerges from the plastic syringe lock, we bend the wire and hold it in place with aluminum tape.


Figure: Electrode H. The Pt-In wire is on the left, EEG pick-up lead enters from the bottom, and a PVC-insulated copper wire passes through a blunt syringe needle.

The implanter can hold the electrode by the syringe lock, and so manipulate it precisely during implantation. Cement will cover the solder joint, but not the tip of the syringe needle. After the cement has set, the implanter cuts the bend in the wire within the plastic syringe lock, and pulls the syringe needle away, leaving a length of copper wire behind.


Figure: Close-Up of Electrode H. The teflon insulation of the Pt-In wire is just discernable around the wire where it meets the solder joint.

The implanter cuts the copper wire off flush with the cement, and covers with another layer to insulate the exposed copper.

[08-SEP-15] We have B85.9, B86.1-8 after three-day soak in water. Battery voltages from average signal value are 2.47-2.53V. Frequency response is within 2 dB of nominal. We have old transmitters A64.1, 3, 5, 7. Battery voltage for A64.1, 5, 7 are around 2.6 V, but A64.3 won't turn on. Reception and mains hum pick-up. Gain versus frequency for both channels of A64.1, 5, 7 all within 2 dB of nominal. Transmitter F66.7 battery voltage 2.5 V, gain versus frequency within 2 dB of nominal, bandwidth 0.3-320 Hz. We open the A64.3 up and find the battery is drained. Current consumption with 2.9 V supply is 1.8 μV when asleep and 149 μV when active. We get perfect

[11-SEP-15] We take an A3028AV4 circuit with its 0.3-160 Hz amplifier and program it to transmit 128 SPS on both channels. We encapsulate in epoxy without a battery, with the help of a crude rotator. We call this device TX1.1


Figure: A Transmitter for Reliability Tests. The motor rotates the circuit as the epoxy cures. We have used vacuum to extract air. The lamp heats the epoxy to accelerate curing.

Transmitter TX1.1 has life 950 hours with a 48 mA-hr battery. If ten transmitters survive in 100% humidity at 60°C for 950 hours with such a battery, we figure their chance of failure in one year at 37°C will be less than 10%.

[15-SEP-15] Transmitter TX1.1 below is encapsulated in epoxy only.


Figure: Epoxy-Only Encapsulation.

We deliver power through a microammeter. We measure current consumption as we lower the device into water all the way up to the programming extension.


Figure: Epoxy-Only TX1.1 Current Consumption with Immersion.

The encapsulation appears to be water-proof. We enclose the transmitter and leads in water inside the finger of a latex glove, and tighten the seal around the programming extension with some silicone tube. We connect a 48-mA-hr battery to the extension, turn the transmitter on, and place in a faraday enclosure jar inside our 60°C oven to poach.

[22-SEP-15] We have batch 1176B consisting of B86.4-5, B86.10-14, B87.2-3, B87.6. Frequency response is within 2 dB of nominal. Reception perfect. Below is reception, battery voltage, and noise for all of them running in water on an antenna.

ID RECEPTION VA NOISE
2 96.29 2.51 10.16
3 99.22 2.52 8.24
4 98.63 2.53 9.28
5 99.02 2.56 8.16
6 98.63 2.52 9.92
10 97.46 2.54 9.52
11 93.16 2.52 7.84
12 98.05 2.51 9.32
13 96.09 2.51 11.68
14 90.23 2.53 9.88

We have E87.7 with water-soluble flux on its components. We turn on with external battery. We observe a 1-Hz full-scale square wave. We wash and blow dry. No more square wave, frequency response within 2 dB of nominal.

We remove TX1.1 from the oven. Our latex water reservoir still contains water, but the jar outside has water condensation all around the walls. There is corrosion between the pins of P2. The battery is drained. We clean and dry the circuit. Current consumption is 2.1 μA when inactive and 54 μA when active. Reception is good, frequency response correct. The electrode solder joints are corroded in a way we do not observe when we poach fully-encapsulated transmitters, see here. The gold pins, gold pads, steel screw, and steel wires show no signs of corrosion. The epoxy encapsulation appears to be unaffected by the poach.

We place the circuit in a zip-lock plastic bag, within a beaker wrapped in foil. We place the apparatus in our oven and tape steel mesh over the window to stop signal getting out. We leave the circuit to poach.


Figure: Epoxy-Only TX1.1 in Water Reservoir for Poach.

The above arrangement keeps TX1.1 in 100% humidity, but as the water evaporates and escapes through the bag, the device will not be immersed in water. Nevertheless, we trust that the saturated environment will be similar to the one encountered by epoxy coated with silicone inside an animal.

[25-SEP-15] We remove TX1.1 from the oven and blow it dry. Inactive current is 1.9 μA. We switch on and current rises as shown below.


Figure: TX1.1 Switch-Once Current Evolution after Ten-Day Poach. Active current after seven days was 54 μA.

[28-SEP-15] Transmitter TX1.1 operating current is 66.5 μA after one minute settling. Gain versus frequency is within 2 dB of nominal (note that this circuit has low-pass filter 160 Hz but sample rate only 128 SPS so we see aliasing at 64 Hz. Battery voltage is 2.66 V with the BR1225 we have been using in the oven.

[29-SEP-15] At first, TX1.1's operating current is 75μA. We scrub, wash, and dry its programming extension. Operating current is now 57.0 μA after one minute, and sleep current is 1.9 μA.

[30-SEP-15] Batch 1177, consisting of E87.7-88.5 made with A3028AV4 circuit, has been through 8hr/60C/D/ON and 40hr/20C/W/OFF. Gain versus frequency is within 2 dB of nominal. Noise is less than 12 μV. Reception is perfect. Battery voltages 2.55-2.66 V. Transmitter TX1.1's operating current is 55 μA immediately after activation, but rises to 57 μA after a few minutes.

OCT-15

[02-OCT-15] Batch 1181 consists of E89.1-14 made with A3028AV4 circuits. We turn them all on and place them in a box on an antenna to pick up mains hum. We measure battery voltage, reception, and signal amplitude. We place in a faraday enclosure with the lid off and do the same measurement. We put the lid on and do the same again.

LocationReception (%)VB (V)Amplitude (counts)
Bench91.52.752228
Enclosure, Lid Off94.42.7642
Enclosure, Lid On92.92.7932
Table: Average Measurements for Batch E89.1-14.

We place the batch in the oven to burn in at 9:36 am. We have R88.7-10 made with A3028RV1 circuits. We turn these on and reception, battery voltage, and noise look fine. We put them in the oven to burn in as well.

Transmitter TX1.1 operating current is 64.3 μA immediately after activation, rising to 66.5 μA after five minutes. Inactive current is 1.9 μA. Transmitter E88.5 is fully encapsulated, made from the A3028AV4 circuit and has been running in 60°C water since 12 pm 30-SEP-15. Battery voltage is 2.70 V.

[06-OCT-15] Transmitter TX1.1's BR1225 battery is drained to 0.5 V. It has been running in water since 15-SEP-15. We last checked on it 02-OCT-15. Operating current is 61.1 μA five minutes after turning on. When we connect a new battery, the average value of X is 42.6k and of Y is 37.2k. Frequency response within 2 dB of nominal. We connect a 1000 mA-hr battery and put TX1.1 back in the oven to continue poaching. A few hours later, we measure average X 42.5k and Y 39.9k. E88.5 running fine, battery voltage 2.7 V.

Batch 1403R R88.7-88.10 have done burn-in and soak, all four are fine. These are replacements for job 1403 failures. Batch 1181 E89.1-14 bave done burn-in and soak. All fourteen are fine: frequency response, battery voltage, reception, and noise.

We have R68.8, R68.9, and R69.3 returned from Marburg. R68.8 consumes over 2 Amps, and C2 starts to smoke. There is green residue around the ends of C2 and C5. We remove C2 and C5. Current consumption is 55 μA. We get no reception. We attempt to replace C2 and C5 but tear off a pad and abandon the effort. R68.9 battery is drained. When on, the circuit consumes only 53 μA. We get no reception. We replace C2 and C5. Current is 70 μA and we see mains hum transmission. After a few minutes, current consumption drops to 52 μA and transmission stops. R69.3's battery is drained. We connect external power. Current consumption is 86 μA. Frequency response is within 2 dB of nominal. Reception perfect.

[09-OCT-15] We remove TX1.1 from the oven. It has drained its 1000 mA-hr battery. Current consumption is 1.8 μA when inactive. When active, current starts at 17 mA and drops after a few minutes to 11 mA. We remove C4, C1, and C3. Active current remains 11 mA. The short circuit is something other than a capacitor, and became permanent after 25 days immersed in water with epoxy-only encapsulation. Transmitter E88.5 continues to run in water at 60°C. Battery voltage is 2.7 V, noise 11 μV, frequency response within 2 dB of nominal.

We add to our poach test two further A3028E transmitter made with the AV4 circuit. They are E89.1 and E89.2. We use our new Function Generator (A3031) and Function_Generator Tool to measure and record the frequency response of both devices, and E88.5 as well.

We receive recordings made from an A3028A-HCC at Edinburgh University. Their mouse recordings are now free of the sudden step artifacts they were observing with their earlier depth electrode assemblies. The trace below is a typical recording from the depth electrode and a screw.


Figure: Typical Recording from A3028A-HCC. The pink trace is the Pt-Ir electrode, the blue is a screw. The common electrode is also a screw. Two-second interval, full scale is 2 mV.

Instead of the sharp step artifacts, we see bumps in the data, like the one shown below. It's not clear to us if these are neurological or movement artifact of some sort.


Figure: Bump Artifact from A3028A-HCC. The pink trace is the Pt-Ir electrode, the blue is a screw. The common electrode is also a screw. Two-second interval, full scale is 2 mV.

These bumps we can distinguish from seizures and oscillations, so we consider these recordings to be a success.

[13-OCT-15] We take E89.1, E89.2, and E88.5 out of the oven. Battery voltages are all three exactly 2.71 V. Frequency response within 1 dB of previous measurements.

[16-OCT-15] Transmitter E90.9 failed during 24/60°C/D/ON burn-in. This transmitter is built with the A3028AV3 circuit. We dissect the transmitter. Battery voltage is 0.8 V. Inactive current is 1.6 μA. Active current is 2.5 mA. Reception perfect. We pick up mains hum. We remove C2 and C5. Active current is 160 mA. We remove C6 and C4, but active current remains 160 mA. We damaged something. Original symptom consistent with failure of 10-μF capacitor C6.

Transmitters E88.5, E89.1, and E89.2 still running. Reception perfect for all. Battery voltages 2.68V, 2.73V, and 2.29V respectively. Frequency response within 2 dB of nominal. We are about to put them back in the oven when we note that E88.5 is transmitting only value 65535 at 512 SPS. We can get it to transmit other values if we shake it around. We turn them all on and put them back in the oven to poach. Plot E88.5 below shows E88.5's frequency response over the course of two weeks of poaching.

[20-OCT-15] Transmitter E88.5 no longer transmits. This device was performing well until it failed during handling on day 17. Battery voltage 0.8 V. Inactive current 1.7 μA. Active current erratic, starting at 3 mA, fluctuating, dropping as low as 150 μA on occasion. Reception is good. With leads connected, average signal value is correct. Remove C6, C4, C5, C3 and current still fluctuates. Transmitters E89.1 and E89.2 performing perfectly. Graphs of their frequency response versus time are E89.1 and E89.2.

Batch E90.1-14 has been burned in for 24 hours and soaked for 4 days. Transmitter E90.9 failed during burn-in. Transmitter E90.14 has a fine 80-Hz cut-off frequency, see E90.14, which means we must have used an 80-Hz EEG amplifier circuit by mistake. All others, gain within &plusnm; 1 dB range, as shown below. battery voltage from 2.58-2.71, noise below 12 μV, reception perfect.


Figure: Gain versus Frequency for Batch E90. All but E90.9 and E90.14 are plotted with Recorder color coding.

We add E90.14 to our reliability tests, turning it on and placing it in the oven to poach with E89.1 and E89.2.

[23-OCT-15] We take out E89.1 and E89.2. We notice E89.2 average value is high, while E89.1 is correct. Soon after E89.2's average value is correct. We obtain fine frequency response from E89.1 and E89.2. After a few minutes, E89.1 and E89.2 average values start to jump around. We dissect E89.1. Battery voltage 2.8 V. Active current is 1.5 mA. We remove C6, C4, and then C5. After removing C5, active current drops to 80 μA, but later jumps up to 6 mA. Inactive current is 1.6 μA. We load fresh capacitors in place of C6, C4, and C5. Current consumption is now stable at 100 μA. We wash and dry, re-connect battery. Average value of X is stable at 42 kcount. Frequency response within 2 dB of nominal. We dissect E89.2. Battery voltage is 2.7 V. Active current 80 μA. We apply 2.6 V, but average value of X is 53 kcounts, implying battery voltage of 2.2 V. We replace capacitor C5 and average value is now 45 kcounts. We measure the resistance of the original C5 after wash and dry, and find it to be at least 40 MΩ. Re-connect battery to E89.2 and frequency response is within 2 dB of nominal.

Comparison of V3 and V4 Capacitors: Transmitters E89.1 and E89.2 suffered capacitor failure after 14 days poaching, although they were still transmitting. We suspect that E88.5 suffered capacitor failure also. TX1.1 failed by some other means after 25 days, but was encapsulated only in epoxy. The 10-μF, 10-V, P0402, general-purpose capacitor by Samsung (CL05A106MP5NUNC, $0.30 each) we used in the V3 circuits survived at least 35 days with rat batteries (8 devices), and at least 9 days with mouse batteries (17 devices). The 2.2-μF, 10-V, P0402, automotive capacitor by Taiyo Yuden (LMK105ABJ225MVHF, $0.15 each) provides a minimum of 14 days of operating life for any battery before any sign of damage (3 devices). All three of these devices failed immediately after we transferred them from water at 60°C to cold tap water at roughly 15°C. In Cracks: The Hidden Defect we read, "An assembly should be allowed to cool to less than 6O¡C before it is subjected to the cleaning process." Epoxy has a thermal expansion coefficient of 50 ppm/°C, while the ceramic dielectric used in capacitors has expansion coefficient a little under 10 ppm/°C. If the epoxy cools first by 40°C, we will have a 0.2% strain upon the outside of the capacitor. Until now, we have not kept track of or controlled the way we cool transmitters when we remove them from the oven to test them.

Alternate Capacitor: The 2.2 μF, 35 V, P0402, general-purpose capacitor by TDK (C1005JB1V225K050BC $0.14). The higher voltage rating we assume implies a larger gap between the plates, which may give us more immunity to corrosion.

[25-OCT-15] We remove E90.14 from the oven and let it sit in its hot water in a faraday enclosure. Reception is intermittent (center frequency drops by 0.4 MHz/°C, so at 60°C center frequency is around 900 MHz, too low for reliable reception). But we see enough to measure battery voltage 2.8 V and noise 6 μV.

[27-OCT-15] Transmitter E90.14 VB = 2.66 V, noise 7 μV, reception 100%.

[30-OCT-15] We have batch E92.1-14 after 24hr/60°C/D/ON. Check battery voltages and reception. All good except E92.8, which has average signal value 49k, implying battery voltage 2.4 V. We remove silicone and epoxy to measure battery voltage, and find 2.66 V. A few minutes later, average value is 52k, but we measure 2.62V. We disconnect battery and measure active current consumption 86 μA, inactive 1.7 μA. If the input offset voltage of U5 is 10 mV on each op-amp, we could have an offset as great as 10 + 2.5 × 10 = 35 mV at U5-1. The average signal value would be 1.835 V, and battery voltage would appear to be 0.05 V too low. Transmitter E90.14 VB = 2.79 V, noise 6 μV, reception 100%.

NOV-15

[04-NOV-15] Transmitter E90.14 VB = 2.79 V, reception 100%, frequency response within ±0.2 dB of 20-OCT-15 measurements in the pass-band 1-80 Hz. We have batch E92.1-14, minut E92.8. Reception is 100% for all. We now find that E92.6 has average value 2.5 kcounts. When we apply large mains hum, we see 60 Hz in X, but we cannot measure a frequency sweep. Reception is 100%. For the remaining 12 devices, frequency response within ±0.8 dB as a group, see here. We dissect E92.6 and find that VCOM is around 0.2 V. Soon after removing epoxy around C6, VCOM jumps to 1.8 V. Current consumption around 90 μA. We did not measure current consumption before we started heating C6. We take out E92.8. We have VCOM = 1.8 V, but U5-7 quiescent value is 2.08, while a 10-MΩ probe grounded to 0V records 1.84 V on U5-6. The triangle wave on VA due to sampling is 34 mVpp. We connect ground of our probe to VCOM on C6 and measure voltages with respect to VCOM. At U5-5 we have 0 mV. At U5-6 we have 4 mV. At U5-7 we have 2000 mV. The offset voltage of U5 appears to be 5 mV, while C6 must be short-circuit. We remove C8 and average voltage drops to normal for VB = 2.6 V. We load a fresh 2.2-μF capacitor in C8. We wash and dry. Average voltage on U5-7 is now 6 mV and on U5-1 is 25 mV. We measure VA = 2.56 V. We expect average X to be 65536 × 1.825 / 2.56 = 46.7 kcounts. We observe 46.4 kcounts.

We have batch E93.1-14 not encapsulated. We connect external 2.6-V power and measure TCK period, center frequency, average X (put leads in beaker of water to suppress mains hum), active and inactive current consumption. All are fine.

We have R77.9 returned from the field, where it failed before implantation. Battery voltage is 0.6 V. We destroy the pad around the 0V battery connection, so further diagnosis impossible.

We work on the P3028A09 firmware, after encountering an AV4 circuit that would give TCK period 193 ns or 220 ns, but nothing in between. In the firmware, we take the fck_divisor constant and use it to configure the length of the ring oscillator and the oscillator divisor. The TCK period is proportional to the product of the length and divisor. In the AV4, the constant of proportionality is 9.3 ns, which is two internal gate delays. We attempt to pick length and divisor so that their product is equal to fck_divisor, but we cannot do this when fck_divisor is prime, nor do we support lengths greater than 11 or less than 2. To get a period of 200 ns, we need fck_divisor 21 or 22. We get 21 with divisor/length 3/7 or 7/3, which both give 194 ns. We get 22 with 2/11, which is why we support ring lengths up to 11. With the 11-gate ring oscillator, the code takes 57 of the available 64 outputs in the logic chip. We obtain this graph of TCK period versus fck_divisor shown below.


Figure: TCK Period versus fck_divisor in Firmware Versions 2, 7, 9, and 11. The period should lie in the range 195-220 ns for perfect reception. The FV11 data we recorded on 29-JAN-16.

The troublesome AV4 circuit now gives 194, 202, and 223 ns period for fck_divisor 21, 22, 24. The 11-gate ring oscillator provides the division by 22, which sets the period between the values 194 and 223 ns, which we obtained in the V7 firmware with divisor 7×3 and 8×3 respectively.

[06-NOV-15] Transmitter E90.14 VB = 2.76 V half an hour after being removed from the oven, but still in its warm water, with noise 6 μV and reception 100%.

[10-NOV-15] Transmitter E90.14 VB = 2.3 V after cooling down. Gain versus frequency 0.8 dB higher, although we do see signs of saturation on the top side of the sinusoidal waves. Reception perfect. Noise 20 μV as signal drifts. Once it's at 60°C again in water, battery voltage appears to be 2.5 V.

We set up the apparatus shown below. Transmitter R76.13 sits in a beaker of water. It is running. We apply 2-V 20-ms pulses at 10 Hz to its input. We cover the beaker with a latex glove. In one finger of the glove is a 10-Ω power resistor. We apply 10 V to the resistor.


Figure: Pulsed Poach Test. This is an attempt to duplicate failure of three transmitters at Marburg when they were implanted with the X input within 1 mm of their bipolar 2-V stimulation electrode.

The beaker will lose most of its heat by radiation, and water is a opaque at infra-red wavelengths. We calculate that 10 W of heat from the resistor should lead to equilibrium at about 60°C. An A3028R should run for 5 weeks at 60°C in water. We will see if the device fails earlier than it should. We set up a computer to record the signal from the transmitter.

[16-NOV-15] Transmitter E90.14 battery voltage 2.6 V, 100% reception after fifteen minutes cooling in its water. We end the R76.13 pulsed poach test. Lots of rust in the water from the alligator clips, and water got through the latex glove to the heating resistor. The water is at 38°C. The transmitter is running perfectly. We turn it off and consider what to do next.

[17-NOV-15] Transmitter B71.7 would not turn on before implantation at ION. The internal circuit is A3028AV3. We perform autopsy. Battery voltage 200 mV. Disconnect battery and supply 2.6 V, active current 160 mA. Clear epoxy from around C2 and C5 (see here), 154 mA. Remove C5, 153 mA. Replace C5, 118 mA. Clear epoxy around C4, 124 mA. Remove C4, 137 mA. Replace C4, 137 mA. Remove C3, 135 mA. This failure is not a capacitor.

We have this this plot of the evolution of the frequency response of E90.14 over the past 27 days poaching. The gain is dropping because the average value of X is around 57k during the sweep. The apparent battery voltage is 2.1 V, which is too low for the circuit to operate. We get 100% reception and 10 μV noise.

We receive 50 of A3028AV5 circuits assembled, as S3028E, with X amplifier 0.3-160 Hz and Y amplifier 0.3-80 Hz. These are made with the MY600 solder paste printer and a covalent-solvent wash (or no-clean process, as they call it.) Our hope is that there will be no ionic residue under U1 and U3 with this process. The big capacitors are 10 μF, 6.3 V, 20% by TDK (445-8920-1-ND). We measure the following frequency response for both channels.


Figure: Frequency Response of X (blue) and Y (orange) in A3028AV5.

We prepare firmware P3028A10, in which device version 8 configures the circuit to transmit on one channel using the Y amplifier. We will use this firmware to make A3028C circuits. The EEG leads will be yellow and blue.

Batch E93.1-14 has endured 24hr/60C/D/ON and 96hr/20C/W/OFF. Reception is 100% from all devices, battery voltages 2.62-2.70V. Gain versus frequency as shown here. We take E93.2 and E93.9, turn them on, and put them in the oven to poach until failure.

[20-NOV-15] We remove our three poaching transmitters from the oven, but do not take them out of their water. Transmitter E90.14 has apparent battery voltage 2.2 V, but is running well. Transmitters E93.2 and E93.9 have apparent battery voltage 2.7 V.

[23-NOV-15] We take out Transmitter E90.14 but get no reception from it at all (AV3 circuit, failure observed on day 34). Transmitters E93.2 and E93.9 running well, battery voltages 2.78 V and 2.81 V respectively.

[24-NOV-15] Transmitter E90.14 was sitting out for 24 hours and was running when we came in this morning. We put it in our faraday enclosure and get 100% reception. Average value is 53 kcount (apparent VB = 2.2 V). Frequency response is within 1 dB of nominal, with top side of sinusoid compressed. Device switches on and off easily. Detects mains hum. Transmitters E93.2 and E93.9 still giving 100% reception. We remove E93.2. Average value 42.3 kcounts, gain versus frequency 0.8 dB lower than a week ago. The transmitter has been sitting out in air for a few minutes, and now shows jumps in average value, example below.


Figure: Cool-Down Jumps In Baseline Value for E93.2, an AV5 Circuit. Interval 1 s, full scale 27 mV.

Transmitter E93.9 has average value 43.0 counts, gain 0.6 dB lower than a week ago. It does not produce jumps in average signal value. We put the three transmitter back in the hot water and watch their signals. E90.14 is stable at 50 kcount. E93.9 stable at 43 kcount. E93.2 spending most of its time at 43 kcounts, but jumping up to 51 kcounts frequently. Six hours later, average values are stable, with apparent battery voltage 2.87 and 2.81 V for No2 and 9 respectively, and noise around 10 μV.

Transmitter E90.14 is still running but battery voltage appears to be 1.98 V and reception is poor. We dissect E90.14. Battery voltage is 2.76 V. Active current 78.5 μA. Center frequency 912 MHz. Average value 56 kcounts. By now, active current has dropped to 74.1 μA. We measure VA on C5 to be 2.0 V. We are supplying 2.7 V from a power supply. We measure VA on R4 and see variation from 2.0-2.7 V. The average value of the signal varies in sympathy. We are pushing on R4 and C5 when the average value suddenly stabilizes at 43.7 kcounts, noise is 10 μV. Reception is 100%. Current is now 91 μA. We measure VA = VD = 2.7 V and 1V8 = VC = 1.8 V on C4 and C6 respectively. We turn our supply voltage down to 1.85 V and quiescent current drops to 75 μA while average value rises to 61 kcounts. We solder the battery back to the circuit board. We have replaced nothing on the board. It now functions perfectly.

There are two ways for VA to drop 0.7 V below VB. One is for the VA current to be 700 μA so the drop across R4 will be 0.7 V. Another is for one of the two U1 mosfet switches to be partially off. We drive U1-5 with a logic chip. But U1-2 is held to 0V with 10 MΩ. We suggest that condensation is compromising the isolation of gate U1-2, causing a 10-MΩ resistance between the gate and nearby source U1-1.

We remove E93.2 from the oven and drop into cold water, then place in faraday enclosure. Average value is 43 kcounts. Ater a few minutes, we start to see sudden jumps in average value, settling to a baseline of 48 kcounts. Jumps are 10 kcounts in amplitude. We record to disk and obtain this overview of 1000 s from M1448397508.ndf, during which the average value settles back to a stable value of 43.8 kcounts (VA = 2.7 V). The behavior of E89.2 is consistent with that of E90.14 and E93.2: jumps in average value without increased current drain. We will call this the "resistive switch" problem.

We have B91.1, 8, 10, and 12 encapsulated with vacuum dip, two drips off the transmitter during 60 s, then 30 s invert to let epoxy flow back, rotate, another coat of epoxy on the bottom side of the board (which we call the "top-coat" for some reason), and five coats of silicone. Dimensions 8.3 mm × 13.8 mm × 14.5 mm (1.7 ml block volume). Also, B91.2, 3, 6, 9 one drip off transmitter during 30 s, then 30 s invert, five coats silicone, 9.0 mm × 14.4 mm × 15.4 mm (2.0 ml block volume). These have the support wire for the rotator clipped off but still visible through the silicone. We measure hand-made B76.10 as 8.3 mm × 13.7 mm × 14.3 mm (1.6 ml block volume). In each case, we take the average of multiple measurements of each dimension around the edges. The B91 circuits are all A3028AV3. We measure frequency response, see here. We turn all the B91 devices on and put them in the oven for 24hr/60°C/D/ON burn-in.

[25-NOV-15] Batch B91 has survived burn-in, we get perfect reception, battery voltages 2.68-2.78 V, and noise around 10 μV. We put them all in the oven to poach.

Transmitter E93.9 is running fine with VA = 2.7 V, but we get no signal from E93.2. We put it in the oven to dry out. Two hours later we still cannot turn on the device. Battery voltage 2.57 V. Disconnect battery. Connect external 2.6 V. Current is 150 mA. We turn off and clean up solder joints. Reconnect power. Active current 90 μA. Now getting reception. Re-connect battery. Reception 100% in faraday enclosure. Gain versus frequency is within 0.1 dB of our measurement pre-poach on 17-NOV-15. Battery voltage 2.61 V, according to average signal value of 46.2 kcount, VA = 2.55 V. Noise is 50 μV. Clear epoxy from around C2, C5, and R4. Transmission stops. Battery voltage 2.7 V. VD = VA = 0.3±0.1 V. VB at U1-1 is 2.7 V. Turn off with magnet and VD = VA = 0.0 V. U1-2 is at 2.7 V. U1-6 = U1-4 = U1-3 = 0.3 V. We connect U1-2 to 0V with ammeter and measure 3 mA. Resistance between U1-2 and VB is 0.0 Ω. We remove U1 with the help of aqueous flux. Don't see corrosion underneath. Wash and dry. Resistance U1-2 to VB is 15 MΩ. We compare the underside of the U1 we removed to a fresh U1. We see no difference. On the footprint of U1 we see epoxy film across the lower of two vias, but not across the upper via, which connects U1-2 to an inner layer. The upper via is not filled with epoxy, and the surface between the via and U1-1 is uncovered.

If we cool a transmitter that is saturated with water vapor, and the U1-2 via is not sealed, condensation in this region will connect U1-1 to U1-2, turning off the mosfet, which will lead to VA dropping. In heat, we speculate that this water will cause corrosion and create a short circuit that will turn off the transmitter. This is our best guess as to the source of the resistive switch problem.

[30-NOV-15] Transmitter E93.9 still going strong, picks up mains hum, and battery voltage 2.7 V. Of the eight transmitters of Batch B91, all are running well and picking up mains hum with battery voltage around 2.7 V, except B91.3, which won't turn on. We put it in the oven to dry out, return all the rest to water to continue poach.

DEC-15

[01-DEC-15] Transmitter B91.3 still does not turn on. We photograph the cut-off wire, which is rusted, as it is in the other three with the cut-off wire covered only with silicone. We see thin cover on peripheral capacitors, as shown below.


Figure: B91.3 Cut-Off Wire End. Transmitter made with rotation-only epoxy, followed by five coats of silicone.

We dissect B91.3. Silicone well-adhered to epoxy. We measure the thickness of the five coats of silicone in various places as we remove it. Estimate the average thickness to be 0.6 mm, consistent with the MED10-6607 data sheet, which specifies 5 mils per coat. We try to solder to the end of the cut-off rotation support lead, after removing silicone, but the lead end refuses to solder until we apply acid flux. Battery voltage 1.2 V. Detach battery and connect 2.6 V. Inactive current 1.7 μA. Active current 45 mA. We remove C5, C4, C6, and C3. Active current remains 45 mA. If we have a short circuit created by corrosion between two power supply pins, we will get such a drain. If the short occurs between U1-1 and U1-2 we get the resistive switch problem. We are diagnosing this as an "Unidentified Drain".

For B91.1, 2, 6, 8, 9, and 12 battery voltage is 2.55-2.60 V. Noise is 10 μV except in B91.12, where we see 20 μV. The noise in B91.12 turns out to be 22-Hz switching noise, fundamental 10 μV, 2nd harmonic 8 μV, 3rd 4 μV, 4th 4 μV.

Transmitter E93.9 has been running fine, but after cooling down, shows steps in its average value, as in the resistive switch problem. We put it back in the oven to poach. After a few hours we take it out again and VA is stable at 2.6 V. Gain versus frequency within 0.1 db of measurement on 17-NOV-15.

[04-DEC-15] We have batch C94.1-8, 0.3-80 Hz using Y in the A3028AV5 assembly. The Y channel amplifier in this assembly is set to 0.3-80 Hz. Gain versus frequency shown here are all within ±0.5 dB. Switching noise 4 μV or less. Battery voltage 2.52-2.65 V. Reception 100%. Total noise less 6-8 μV.

Of the transmitters poaching, E93.9, B91.6, and B91.10 no longer transmit. The remainder, B91.1, 2, 8, 9, and 12 have battery voltages 2.48-2.64 V. Reception is perfect and noise is less than 10 μV.

After an hour sitting in air on our bench, B91.10 transmits. We get poor reception in faraday enclosure, VA = 1.9 V. Dissect. Silicone is so well adhered to epoxy that we must shave it off. Battery voltage 2.74 V. Inactive current 1.6 μA, active 62 μA no reception. Apply heat gun to bottom side, active current 72 μA some reception. Heat U1 up with solder blob for 20 s. Wash and dry. Active current 90 μA for thirty second, reception perfect. VA = 2.7 V. Current drops down to 60 μV and we lose reception. Replace R2 with 0Ω. Active current 100 μA, VA matches our applied VB. After thirty seconds, VM drops to 1.0 V, current jumps to 3.7 mA, and VB has dropped to 1.0 V because of he impedance of our ammeter. We see U1-2 at 0V. We switch to milliammeter and current jumps up to tens of milliamps then drops to 100 μA. After thirty seconds, jumps up to 4 mA and reception continues. We turn off and on again. Current is 90 μA and remains so for twenty minutes. Remove U1. Note that both vias under the chip are filled with black epoxy.

We dissect E93.9. Silicone comes off easily with the black varnish. Battery voltage 2.5 V. Current consumption jumps up to 3 mA briefly. When 80 μA get good reception. After one minute, current settles to 80 μA. We see no aberrations for twenty minutes.

We dissect B91.6. Silicone is well-adhered to epoxy. Battery voltage before disconnecting is 0.5 V, after disconnecting 1.0 V. Connect 2.6 V. Inactive current 25 mA. Remove glue from around C2. Inactive current 21 mA. Remove C2. Inactive current 21 mA, erratic. We have VB = 2.7 V, VA = 0.5 V, VD = 0.5 V, VM = 2.7 V. Heat up epoxy around U1 with intent to measure voltages on it. Current drops to 10 μA, VM = 2.7 V, VD = 0.5 V. U1-5 = U1-4 = VM. We see U2-2 switching with magnet. But the circuit does not switch on. We see U3 is displaced on its footprint.

The failure of B91.10 is consistent with the resistive switch problem. The Failure of E93.9 and E91.6 are appear to be due to a short between U1-5 and U1-4, which results in VD being turned off, and current being drained through U3. We observed a short between U1-1 and U1-2 before, so U1-5 and U1-4 could also be shorted by corrosion. But we were unable to measure this short directly today. We classify this problem as a "corrosion short".

[07-DEC-15] Of the 5 transmitters poaching, B91.1, 2, 9, and 12 are running well, with battery voltages 2.58-2.69 V. But B91.8 does not transmit. (See 08-DEC-15 below for possible explanation: we had B91.8 and B91.9 mixed up.) We leave it out on our work bench to dry and put the others back. When we return to our bench, B91.8 is running. Battery voltage 2.61 V. Gain versus frequency has changed: the bump at 130 Hz is gone, see here. Put back in the oven to poach.

[08-DEC-15] We have batch C94.1-8 after four days soaking in water. All detect heartbeat. Battery voltages 2.57-2.70. Noise less than 12 μV. Reception perfect. This is the first batch of circuits made with the A3028AV5 circuit. We put C94.3 and 4 in the oven to poach, and ship the rest to ION.

Of B91.1, 2, 8, and 12 are running well. Battery voltages 2.4, 2.62, 2.62, 2.50 V respectively. Noise less than 12 μV. Reception perfect. We measure gain versus frequency and plot B91.1, B91.2, B91.8, B91.12. B91.1 and B91.8 show gain 3 dB too low, while the others are within 0.1 dB of their original gain. B91.9 does not run at all. We suspect that B91.9 was not running yesterday morning either, but in our haste, and with an Octal Data Receiver with no channel number labels, we mixed up B91.8 and B91.9. We will assume this mix-up took place. We now dissect B91.9. Battery voltage 1.6 V. Solder leads to VB and 0V. Transmitter starts working on its own battery for thirty seconds, turns off. Apply 2.6V. Active current 90 μA. Inactive 1.6 μA. Picks up mains hum, reception perfect. Battery, now disconnected, has recovered to 2.3 V. We connect a fresh battery and measure frequency response B91.9. We are going to call this "Unidentified Drain".

[11-DEC-15] B91.1, 2, and 8 have all stopped. B91.12 is running well, VA = 2.53 V, reception perfect, noise 19 μV. C94.3 and 4 are running well, VA = 2.71 and 2.67 V, reception perfect, noise 7 μV.

We dissect B91.2. Battery voltage 0.0 V. Disconnect battery, 1.0 V. Connect external 2.6 V. Transmitter switches on and off. Inactive current consumption climbs from 20 μA to 4 mA in two minutes, and is still climbing. Transmitter still turns on and off, now with active current 5.07 mA and inactive 5.00 mA. Remove C2. Inactive consumption 1.8 μA, active 66 μA. Reception intermittent. Resistance between C2 terminals is 36 Ω. Clean with solder and alcohol, still 36 Ω. Replace C2 with fresh part. Active current 83 μA. Reception perfect. Connect to a battery. With voltmeter, VB = 3.2 V, with X, VA = 3.2 V. Measure frequency response follows same shape as earlier measurements, but is lower because of elevated battery voltage. Classify as "corroded capacitor".

We dissect B91.8. Battery voltage 1.3 V. Disconnect battery, 2.1 V. Connect external 2.6 V. Inactive 1.7 μA, active 280 μA and climbing. Robust reception. Average X 31k. Active current now 800 μA, average X 25k. Remove C6. Active current 90 μA. Average X 45k. Resistance of removed C6 is ∞. Measure gain, it is farther off than it was a few days ago, B91.8. Classify as "corroded capacitor".

We dissect B91.1. Battery voltage 1.8 V. Disconnect battery, 2.8 V. Connect external 2.6 V. Inactive 1.8 μA. Active 88 μA. Average X 45k. Gain has dropped farther, see B91.1. If we drive X with a 50-Ω source, gain versus frequency is perfect. We classify this failure as "unidentified drain".

We have batch E95, consisting of E94.1-95.7, made from AV5 circuits. Gain versus frequency as shown in E95. All look good except E95.3, with gain far too low at 100 Hz, and E94.12, which has no bump in gain at 120 Hz. Battery voltages 2.54-2.77, reception perfect, noise less than 12 μV. Switching noise less than 4 μV. We fail E94.12 and E95.3 for frequency response and put them in the oven to poach. Ten of the remainder ship today.

We are ordering another 50 AV5 circuits with lead-free solder, no-clean process, and paste printer. These will be AV5LF for lead-free.

[15-DEC-15] We have five devices poaching. We remove them from hot water for approximately three minutes in dry air. C94.3 and C94.4 have robust reception, VA = 2.78 and 2.76 V respectively. B91.12 robust reception, VA = 2.63 V. E95.3 and E95.12 robust reception, VA = 3.05 and 2.78 V respectively. Noise below 12 μV.

[17-DEC-15] Of five devices poaching, C94.3 and C94.4 have robust reception, VA = 2.78 and 2.76 V respectively, which we note is identical to the values we obtained two days ago. B91.12 has stopped transmitting after running continuously for 23 days, 22 of which were at 60°C in water. E94.12 and E95.3 have robust reception, VA = 2.82 and 2.85 respectively, noise below 7 μV.

[18-DEC-15] E94.12 and E95.3 have robust reception, VA = 2.80 and 2.85 V respectively, noise below 7 μV. C94.3 and C94.4 have robust reception, VA = 2.77 and 2.75 V respectively, noise below 7 μV.

[21-DEC-15] C94.3 and C94.4 have robust reception, VA = 2.88 and 2.73 respectively. Noise less than 9 μV. E95.3 and E94.12 have robust reception, VA = 2.84 ad 2.76 V respectively, noise less than 8 μV.

[22-DEC-15] C94.3 reception robust, VA = 2.81 V, noise 6.6 μV. C94.4 reception robust, VA = 2.73 V, noise 9.0 μV. E95.3 reception robust, VA = 2.85 V, noise 6.5 μV. E94.12 not transmitting. We dissect. The silicone comes off the enamel-painted surface easily in one piece, with some of the enamel coming with it. There are three cavities in the epoxy top-coat. Covering over some of the capacitor corners is thin. Battery voltage 2.9 V. Remove C2 and C5. Battery voltage appears on VB. Connect external 2.6 V. Current consumption 0.2 μA. U1-2 varies 2.5±0.4 V when we apply 10-MΩ probe. Resistance U1-2 to 0V is 11 MΩ. Resistance U1-2 to VB is 0.5 MΩ. This is the resistive switch problem. We replace R2 with 100 kΩ. We replace C1 and C5. We wash and blow dry. Inactive current consumption is 26 μA and resistance from U1-2 to VB is now 2 kΩ. We dry in the oven. No change. We replace R2 with 0 Ω. Now inactive current consumption is 1.6 μA, active 60 mA. We have VA = 1.8 V. VM = 2.2±0.1 V. This is a corrosion short problem. We remove U1, R2, C2, C5, clean, and dry. Resistance from U1-2 to VB is ∞ and from U1-5 to VM is 1.7 MΩ.

U1 is directly below the epoxy cavity. We examine batch E97 and find that 6 of 14 have bubbles in the top-coat (the coat of epoxy on the bottom side of the board, over U1, C2, C5). In three cases, there is a bubble over U1, but not in the other three cases. We note that we saw the resistive switch and corrosion short problems in our rotator-encapsulated batch B91, which had no bubbles nor enamel, and silicone was well-adhered to entire surface. These AV5 circuits are made with a water-free process, made with paste printer and leaded solder. Our only hypothesis to explain the problem under U1 is that we are not washing our flux off sufficiently after loading the battery. We started double-washing the circuit after battery loading with batch C95 a couple of weeks ago.

We have batch C95, consisting of A3028C-AA numbers C95.8-C96.5. This is the first batch we double-washed after battery loading. Frequency response is C95_G_vs_F, all within ±0.8 dB at the bump, and ±0.2 dB elsewhere. Reception is robust, battery voltages 2.51-2.60 V, noise less than 12 μV, switching noise less than 4 μV. We take C96.4 and C96.5 for aging tests. We turn them on and put them in the oven to poach.

[23-DEC-15] C94.3 reception robust, VA = 2.80 V, noise 5.3 μV. C94.4 reception robust, VA = 2.71 V, noise 9.3 μV. C96.4 reception robust, VA = 2.81 V, noise 4.9 μV. C96.5 reception robust, VA = 2.77 V, noise 5.8 μV. E95.3 not transmitting.

[24-DEC-15] C94.3 reception is 70%, VA = 2.79 V, noise 5.0 μV. We remove from hot water and allow to cool down. Reception is 100%. Frequency response C94_3_G_vs_F is within 0.2 dB of 20 days ago. C94.4 reception robust, VA = 2.61 V, noise 7.0 μV. C96.4 reception robust, VA = 2.82 V, noise 4.7 μV. C96.5 reception robust, VA = 2.78 V, noise 6.0 μV. E95.3 has been sitting on our bench drying out. Still does not transmit. We dissect. Silicone comes easily off the enamel. Battery voltage 3.0 V. Connect external 2.6 V. Inactive current 0.5 μA. Cannot activate. Clear epoxy around R2. Current 0.5 μA. Measure voltage on U1-2 where is appears on R2, get 2.6 V. Measure resistance U1-2 to VB, get 3.8 kΩ. We clear epoxy from around U2. Resistance U1-2 to VB now 20 MΩ. Connect power, inactive current 1.6 μA, active 85 μA. Measure frequency response C95_3_G_vs_F and note that original problem with response has disappeared. We obtain the following photograph of U1, which shows a dendrite growing from U1-1 towards U1-2.


Figure: Dendrite Growing from U1-1 to U1-2. This circuit suffered from the resistive switch problem until we cleared epoxy from around U1. R2 is on the lower left. Above R2 is C11. At center is U1, with pins 1-3 from left to right. The shiny line is the dendrite.

The dendrite remains shiny regardless of the angle of our light. We remove it with tweezers and photograph it alone. We bend it with our tweezers. We unbend it. It's tiny, but we are certain it is metallic. We note that E94.3 is made with the lead-free AV5 circuit. The dendrite did not form beneath U1, but between its pads. For this to occur, something must have occupied space between the pads to stop epoxy from insulating them from one another. Subsequently, whatever space filler was present took part in migration of Sn or Pb between the two terminals with 2.7 V driving the reaction.

These boards are made with lead-free solder, a solder paste printer, and covalent chemical wash. The U1 package is lead-free, but that's because its copper pins are coated with tin. The circuit boards themselves are gold-plated copper. There should be no silver around U1. The dendrite is made of lead or tin or both. In Choi et al. the authors measure the electromigration activation energy for SnPb solder, and obtain a value of 0.77 eV. The activation energy for silver dendrite formation was measured by Hornung to be 1.1 eV. In this report, they claim silver is much more likely to migrate than tin or lead. But let us suppose the activation energy for SnPb migration is 0.77 eV. In that case, our acceleration factor for 60°C (333 K) compared to rodent body-temperature 37°C (310 K) is exp(0.77/310k −0.77/333k) = 71. For an activation energy of 1.1 eV the factor is 16. Because this resistive switch failure is PbSn, it's likely that onset at 11 days at 60°C implies onset at 780 days at 37°C.

[28-DEC-15] We have batch E96.6-97.3, which we call E96, after 24hr/60°C/D/ON and 72hr/20°C/W/OFF. Reception is perfect from all devices. Gain versus frequency within 1 dB of nominal, see E96_G_vs_F. Switching noise less than 6 μV. VS = 2.57-2.64 V.

We take C94.3 out of 60°C water and let it cool down for one minute. Reception is 100%. VA = 2.54 V. Frequency response C94_3_G_vs_F looks good. We place in 45°C water, RF center frequency 909 MHz. Transmitter C94.4 we remove from 60°C water, frequency response C94_4_G_vs_F. Transmitter C96.4 reception perfect, VA = 2.81 V, noise 5.3 μV. Transmitter C96.5 reception perfect, VA = 2.77 V, noise 5.76 μV. We start testing of transmitters E97.2 and E97.3, both AV5 circuits, both double-washed.

[30-DEC-15] We check 6 transmitters that have been poaching at 60°C. Transmitter C94.3 reception perfect, VA = 2.56 V, noise 8.8 μV. Transmitter C94.4 reception perfect, VA = 2.45 V, noise 8.28 μV. Transmitter E97.2 reception robust, VA = 2.79 V, noise 6.7 μV. Transmitter E97.3 reception robust, VA = 2.81 V, noise 6.6 μV. Transmitter C96.4 reception robust, VA = 2.82 V, noise 4.5 μV. Transmitter C96.5 reception perfect, VA = 2.79 V, noise 5.8 μV.

We have batch E97.4-98.2, AV5 circuits. Gain versus frequency E97_G_vs_F. Noise <10 μV, switching noise less than 4 μV, all 19-21 Hz.