Subcutaneous Transmitter (A3048) Development and Production| 2023 | 2024 |
[30-MAR-23] Start circuit design.
[06-APR-23] We remove the low-pass filter that lies between VD and VA on an A3047A1A. That is: we remove R1 and replace with 0 Ω, leaving the 10-μF capacitor in place on VA. We see no increase in noise, no switching noise. The logic on the A3047 uses only 1V8. The only things using VD and VA are the VCO and the amplifiers. We resolve to remove this resistor from the A3048. We start with A3028PV3 circuit, replace LTC1865L with ADS8860, which we first tried out in the A3047. We now have enough free space to add a two-component antenna matching network. We are unable to fit two SC-353 single op-amps on the board. We must stick with the SOT-23-8 dual op-amp. We cannot purchase the OPA2349. The MAX4471 is available. It is a drop-in replacement with the advantage that it is rail-to-rail input and output, its input offset is only 0.5 mV, and its quiescent current is slightly lower. But its gain-bandwidth is only 9 kHz compared to the OPA2349's 70 kHz. But the only pup transmitters we have been making are 0.3-40 Hz and 0.3-80 Hz. The first amplifier has a gain of 40, so we need 3.2 kHz gain-bandwidth product. The MAX4471 will do the job easily, and will do okay with 160 Hz as well. We rotate the VCO so we can add our matching network. The RF signal propagates diagonally across the circuit board. We exchange the power and ground planes for two additional signal planes, making six signal copper planes in all. Re-name components so they are contiguous.
[08-APR-23] Layout A304801A complete, schematic S3048_1.gif.
[11-APR-23] Printed circuit board submitted for fabrication.
[14-APR-23] Panel Gerber files received. Note that logic chip LC4064ZE-7MN64I now in-stock at DigiKey.
[09-MAY-23] Receive 100 of A3048AV1. These are equipped with 2-nF capacitors for 80-Hz bandwidth, and resistors for ×100 gain. After correcting one constraint error in the code, the circuit works perfectly. It does have an , with the one idiosyncracy: when we power it on through our multimeter set to microamps or milliamps, the resistance of the meter causes the circuit to become stuck in a state where it consumes several milliamps and does not complete its turn-on. When powered by a battery, the circuit always turns on and off correctly. We program and test three circuits at 128 SPS, 256 SPS, and 512 SPS and find the average current consumption is 29, 42, and 69 μA respectively, which is 10% lower than the consumption of the A3028KV2. We compare the saturation behavior of the A3040D2 amplifiers, which are identical to the A3028S2 amplifiers, to that of the A3048S2.
[12-MAY-23] The A3041AV1 amplfier is equipped with a three-pole low-pass filter and MAX4471 9-kHz dual op-amp. If we remove this filter, we see the maximum bandwidth the amplifier can deliver. With 1.0 nF capacitors, the amplifier has a corner frequency of 160 Hz, and with 2.0 nF capacitors the corner frequency is 80 Hz. The AV1 assembly comes with 2.0 nF capacitors by default.
The A3048AV1 can provide a gain of ×100 with corner frequencies of 40 Hz, 80 Hz, and 160 Hz, but no higher. Its amplifier is not fast enough to provide a gain of ×100 at 320 Hz. The A3048BV1 will provide a faster amplifier.
[06-JUN-23] We add another passive component to the A304801A to make a T-network between the VCO and the antenna. We convert the amplifier to provide gain ×21 in first stage and x5 in second stage, for total of x105, as we did in A3049. The OPA2369, with its 12-kHz gain-bandwidth product, will provide gain ×21 up to roughly 500 Hz, while at the same time guaranteeing offset less than 0.75 mV, making the circuit suitable for amplifying biopotentials down to 0.0 Hz. We make some other adjustments to tracks and silk screen, generating A304801BR1, which we submit for fabrication, and new schematic S3048B_1. Assembly BV1 will be equipped with T-network C12=C13=15pF and R14=200Ω. This network gives complete protection against sparks from our plasma ball. It attenuates the transmit power no more than 1 dB. The BV1 will be loaded with 2.0-nF filter capacitors for 80-Hz corner frequency.
[28-JUN-23] We have two A3048S2 that failed in the same way during poaching, each after roughly twenty days. Prior to failure, sweep response is perfect every day. On the day of failure, the device won't turn on. Dissect both. Battery voltage is 1.5 V until we disconnect, then rises to 2.7 V. Connect battery to circuit, voltage drops to 1.5 V again. Jump start by connecting 2.7 V across battery briefly. Circuit powers up and transmits. Disconnect from battery. Connect external 2.7 V. Ater initial burst of current to power up the circuit, current consumption is ≈40 μA. Connect 10 kΩ to battery, voltage drops to 1.5 V. Connect 10 kΩ to fresh battery, voltage remains 3.22 V. If we connect either battery to an A3028KV2 circuit, one that consumes 75 μA, the battery can turn on the circuit, and battery voltage is around 1.9 V.
In a batch of 23 A3048P2, equipped with CR1025 battery, we see switching noise up to 2 μV rms, which consists of pulses of around 20 ms and height up to 30 μV at room temperature. For spectrum see here. We see no sign of such noise in the A3048S2 equipped with the CR1225 battery.
[30-JUN-23] Receive 120 of A3048AV1. Measure current consumption versus sample rate, add to our existing measurements, slope 0.106 μA/SPS, intercept 16.1 μA.
[19-JUL-23] Firmware P3048A05 provides uniform sampling with transmission scatter. The uniform sampling is achieved by always sampling at the end of each sample period, by asserting CSS for one CK period. The active CK period, when we read out the sample, takes place 1 to 16 CK periods later. We assert CSS only during the ADC readout, not for the full CK period. Applied to an AV1 assembly at 256 SPS we have current consumption 43.1 μA with scattered sampling and transmission. We have 43.4 with uniform sampling and scattered transmission. Distortion at 50 Hz drops from 40,000 ppm to 4.3 ppm.
[01-AUG-23] We have 200 of A3048BV1. First problem we discover is we specified 100 kΩ for R3. We must swap for 4.02 kΩ on all boards.
[18-AUG-23] We check the RF power emitted by the A3048AV1, with no antenna protection network, and the A3048BV1, with three-component T-protection network. Nathan reports. "We measured the RF power output of the A3048BV1 and compared it to that of the A3048AV1. We programmed and calibrated both boards. We then placed each one separately in a faraday enclosure and measured its power output using the spectrometer tool. They had comparable power output. We then tested its static protection by shocking the antenna of the transmitter with a spark from a plasma ball with a washer on top. The A3048BV1 survived the shocks from the plasma ball and operated perfectly fine afterward. The 3048AV1 lacked protection and would stop working after a couple sparks. The VCO would need to be replaced, indicated by the 18mA current consumption."
[02-OCT-23] We have our first batch of A3048 transmitters made with the A3048BV1 circuit. They are a batch of A3048S2-AA-C50-D. Noise in 1-80 Hz is 2-3 μV rms, a new record low for a batch of transmitters.
[31-JAN-24] In a batch of 24 A3048P2 transmitters, after one-day soak, we find that we have to turn each one on three times if we are to be certain that the circuit powers up correctly. When it powers up incorrectly, the signal reports value zero always. After QC2, we turn them all off, wait a few minutes, and turn each one on again with one touch of the magnet, every one of them powers up correctly. We are leaving them to soak for another few days, but this incident alarms us enough to discontinue use of the CR1025 battery and replace it with the CR1216, which has the same capacity, volume, and mass, but is provided by Murata with guaranteed pulsed current performance. The CR1216 by Murata can provide 10 mA for 10 ms in starting an A3041 IST. It can certainly provide 2 mA for 1 ms to power up an SCT. We choose version letter "R" for the new line of transmitters.
[09-FEB-24] We have our first batch of 24 A3048R2s passed through QC2, frequency response here. Mass is 1.67 g (nominal is 1.6 g), volume is 0.84 ml (nominal is 0.80 ml). The battery is as thin as the circuit board after encapsulation, see here.
[22-MAR-24] We measure current consumption for falling and rising battery voltage for an A3048BV1, so as to observe the latch-up current of the ADS8860.
Nathan reports, "We measure the current consumption of an A3048S2 transmitter with respect to its battery voltage. First, we start by applying 3V and decreasing the voltage down to zero while measuring current consumption and notice that the device stops transmitting around 1.8 V. Then, we start at 0V and increase the voltage applied on the circuit to 3V noting that the device begins to transmit around 2V." We note that this behavior will cause the transmitter to drain its battery if the power supply rises too slowly, or if it drops slowly. Slow drops are likely at the onset of corrosion in implanted transmitters, and we see such sudden drains in our poaching transmitters.
[25-MAR-24] During QC2 we find that No81 from a batch of A3049J2-AAAA-C45-D turns on for the first time and latches up: signals are both stuck at 65535. Turn off, turn on and it works fine. Turn off and wait ten minutes, turn on again, it latches up. We can get it to latch up or down (stuck at zero). We dissect. Current consumption is normal. We load the Renata CR1225 battery with 5 kΩ and observe source resistance 500 Ω. We load a fresh Renata CR1225 battery with 5 kΩ for a few hours, its source resistance is 100 Ω. We observed when choosing batteries for our ISTS that some Renata batteries have lower source resistance than others. The Renata batteries have no specified behavior for pulsed currents.
[04-APR-24] We have an A3048S2, S234.62, that generated 40 μV, 4-Hz spikes during QC2, this being switching noise from the magnetic sensor. We poached it. It failed 36 days, a few days short of its nominal 41-day operating life. We take this device and use it to investigate the relationship between battery source resistance and switching noise amplitude. Nathan reports. "Attached is an image of my setup for performing an experiment to measure the switching noise in relation to the source resistance in series with a CR2330 battery as the power supply. We increase the source resistance on the battery by adjusting a potentiometer and measuring its resistance each time. We then take a recording and play it back in the neuroplayer to view the fourier transform of the transmitter signal. Measuring the amplitude of the switching noise and its harmonics allows us to plot a relationship. See plot attached as well."
Switching noise harmonic amplitude, as seen in its spectrum, increases linearly with the resistor we insert between the battery and the transmitter.
Above 150 Ω the device latches up when we turn it on. Above 400 Ω it will not turn on at all. In order to obtain our measurements above 150 Ω, we start the circuit with a low-impedance source.
[10-APR-24] We measure the current drawn by A3048 and A3049 circuits during power up. As we report in Startup Current, both circuits consume 7 mA for 5 ms, then 3 mA for 120 ms, when supplied with 2.7 V through a 100-Ω resistor. The first rush is current flowing into the decoupling capacitors. The second rush is consumed by the ADS8860 analog to digital converter.
[16-APR-24] In March, we received S208.5 back from ION. It failed suddenly while implanted. We dissect and find VB = 2.9 V. In the recording of the moment of failure, X jumps 1 mV, then the transmitter turns off. We see the same failure in S237.102 after five days poaching. Nathan reports, "The signal from the transmitter looks like a saw tooth and its current consumption changes with small movements. To investigate this issue, we dissect. The battery voltage appeared normal at about 2.9 V both loaded and unloaded but the current consumption (only when on) was 200 μA higher than expected. This measurement did not decrease with applying 4.2 V to VB nor did it change when I heated the circuit. The current consumption didn't change much when I took off the VCO (U6); it only changed when I removed the ADC. With that in mind I dissected the transmitter that came back to us from Amy and replace its ADC. Once its ADC is replaced, the transmitter behaves normally."
When these two circuits were exhibiting their failure, touching the antenna or body lightly would improve reception for a moment, thus making it possible for us to see the saw-tooth provided by the signal. We see no way a faulty ADC could compromise reception. These transmitters both use the P3048A03 firmware, which latches SDO on the rising edge of TCK. Regardless of what the ADC does, there is no way for the transmitted bit stream to deviate from the format provided by the TXS state machine. For the ADC to compromise reception it must be dropping the VA power supply by consuming excessive current. When we do see reception, the ID is correct and present, but the ADC output appeasr to be either $8000, $0000, or $FFFF.
We wonder if firmware glitches could cause the ADC to misbehave. We note that SCK and CSS are partly combinatorial in the P3048A03 firmwarwe. We prepare and test firmware P3048A04, in which SCK and CSS are generated synchronously with TCK and CK.
We check frequency response (correct), noise (22 cnt rms), and distortion (<20 ppm). Current consumption at 64 SPS is 23.0 μA, at 256 SPS is 43.8 μA, and at 2048 SPS is 243 μA. Slope is 0.11 μA/SPS. Intercept 16 μA.
[29-APR-24] As we report here, the CR1225 from Multicomp provides far superior pulsed-current performance to the Renata CR1225, as well as what appears to be 50 mAhr capacity. We stop use of the Renata battery and start using the Multicomp.
[03-MAY-24] We have a batch of 14 of A3048S2 made with Multicomp CR1225. Noise is ≤2.5 μV rms, a new record. No trace of switching noise. The irregular pulses of around ten or twenty microvolts that we have been seeing in the S2 made with the Renata CR1225 are not present.
[21-MAY-24] Yesterday we loaded a CR927 into an A3048S5 circuit, 2048 SPS. This CR927 had just completed a high-power stimulus endurance test, and was no longer able to power up an IST. Today we find that the transmitter signal shows 1-mV switching noise spikes. Our high-power stimulus endurance test conists of 10-ms, 10-mA pulses drawn from the battery every 100 ms for 900 s, then 100 s rest, then start again. The amplitude of the pulses is around 1500 μV. The first harmonic is at 4.75 Hz with amplitude 80 μV.
[04-JUN-24] We have been trying out ultraviolet light as a way to find flux residue on our circuit boards. Calvin reports. "Kirsten had to fix a couple of leads and tabs on a few S2 transmitters and we took the opportunity to look one last time with the UV light and see if we could see the flux residue. While we were able to make out the flux on both the lead and tab joints, it was not very clear and required the use of a loupe and some careful inspection. In the end it seems like using the UV light is not significantly easier than using the reflection of overhead lights as we do now and probably would require just as much training to be able to do consistently. I think with this I am satisfied that the UV tests can be concluded, it was an interesting idea but probably more trouble than it is worth."
[11-JUN-24] We update our A304801B Traxmaker PCB file to reflect changes made by Epectec at our request to the bottom silkscreen we sent them back on June, 2023. The bottom silkscreen now says A304801BR1 and 06-JUN-23.
[14-JUN-24] The past ten batches of A3048S2 have had average mass 1.9 g, so we are increasing the mass specification for the A3048S2 from 1.8 g to 1.9 g. We revive the P-series transmitters with the CR927 battery.
[19-JUN-24] Our new CR-Series Multicomp and Murata batteries maintain a higher battery voltage throughout their lifetimes compared to the original BR-series Panasonic batteries we used in our transmitters ten years ago. In the past we used 2.7 V as our nominal battery voltage, because the BR-series batteries produced 2.7 V half-way through their life. But our new batteries produce 3.0 V for the first half of their life, so we increase the A3048 nominal battery voltage to 3.0 V. Dynamic ranges increase from 27 mV to 30 mV.
[24-JUN-24] We compare two batteries made by PHD, their CR927 and CR1025. We load with 27 kΩ, which draws a little over 100 μA. Both batteries have nominal capacity 28 mAhr.
All four of our test batteries deliver 28±1 mAhr. The pulsed load performance of the CR1025 is, however, far inferior to that of the CR927, as we show here. We compare the CR1225 batteries manufactured by Multicomp and Renata.
The Multicomp data sheet states 50 mAhr capacity. We see around 56 mAhr. The Renata data sheet states 48 mAhr. We see around 48 mAhr. The pulsed load performance of the Renata CR1225 is, however, far inferior to that of the Multicomp CR1225, as we show . We have been using the Multicomp battery in our A3048S transmitters for the past few months. Today, two of our poaching A3048S2 transmitters failed after 47 days, when their advertised lifetime is only 41 days.
[01-JUL-24] We have two A3049P2 made with PHD CR927 batteries, see A3048P_CR927. Volume 0.70 ml, mass 1.5 g. Compare to the discontinued A3049P made with Renata CR1225 battery, see A3048P_CR1025, volume 0.8 ml, mass 1.6 g.
[03-JUL-24] Nathan reports. We program an A3048 board to be an S2 transmitter and connect it to a benchtop power supply. The benchtop power is supplying 2.7V to our A3048 circuit through a potentiometer and an ammeter on its mA setting. The potentiometer is acting as a linear 200R rheostat in series with the S2 transmitter. We start by gradually increasing the resistance in series with the transmitter until the transmitter fails to start properly regardless of how many times we switch it on or off. We measure the resistance to be 165R and we attach our scope probe to either side of the potentiometer to look at the startup behavior. The resistance in series with our transmitter drops the voltage applied to the circuit by 1V, meaning that 1.7V is barely insufficient for powering the transmitter. This is consistent with our earlier experiment where we gradually changed the voltage applied through benchtop power and noticed that below 1.8V the transmitter would fail to start properly. We also take the same measurement with only 42R in series with the power supply to compare a startup failure to a startup success. Note the difference in vertical scaling between oscilloscope screenshots.
[08-AUG-24] We have four faulty A3048S2 returned from ION/UCL. We dissect to determine cause of failure. All four appear to have suffered from corrosion. Nathan reports in detail below.
S208.3: We measure the loaded battery voltage to be 1.573V and its unloaded battery voltage to be 2.382V indicating that the circuit was consuming more current than usually and dragging down the power supply. We remove the battery and apply benchtop power to measure its current consumption. It consumes 42.9 μA in its on state and 1.0 μA in its off state. This is within the expected current consumption range of an S2 transmitter which tells me that any corrosion that was causing increased current consumption must have been removed during dissection. Diagnosis: Corrosion Drain.
S208.9: We measure the loaded battery voltage to be 27mV and the unloaded voltage to be 35mV. This tells us the battery was fully drained past where a transmitter can no longer power up. Battery voltages wouldn't drop this low unless corrosion caused some kind of short in the power supply and fully drained the battery. We measure its current consumption to be 44.6 μA in its on state and 0.9 μA in its off state. Again, whatever corrosion did occur in the circuit must have been cleared during dissection since it behaves normally with benchtop power. Diagnosis: Corrosion Short.
S208.147: We measure its loaded battery voltage to be 0.5mV and its unloaded voltage to be 160mV. Similar to S208.9, this transmitter had its battery fully drained. When trying to measure its current consumption through benchtop power and an ammeter on its mA setting we overload the ammeter and blow its fuse. This means the transmitter must have been consuming at least hundreds of mA. It consumes only 1.8 μA in its off state. This could only occur when the power supply has a solid short in the transmitter circuit. We dissect the transmitter further, allowing it to heat up and observe its current consumption drop to the standard range for an S2 Transmitter. Diagnosis: Corrosion Short.
S208.150: We measure its loaded battery voltage to be 2.468V and its unloaded voltage to be 2.432V. We notice that when the battery is loaded it still transmits but its signal is saturated to the top rail. When using benchtop power the transmitter consumes roughly 220 μA in its on state and 1.1 μA in its off state. Looking at the receiver instrument while turning on this transmitter shows us sharp saw-like waves. These symptoms are indicative of an ADC failure. We reflow the joints on the ADC and notice the transmitter return to normal. This tells us that the corrosion most likely took place between pins of the ADC and not inside of the ADC. Diagnosis: Corrosion Short across ADC pins.
[09-JUL-24] We drained four CR1225 manufactured by PHD with 27-kΩ load, for roughly 100 μA current drain. These batteries have proved themselves in other tests with pulsed currents.
[16-JUL-24] In our most recent batch of 200 A3048BV1, we notice that the frequency response frequently contains either an excessive bump in gain at 70 Hz, or an excessive dip in gain at 70 Hz. We select three such deviant circuits from our latest batch in production and plot their sweep response.
Capacitors C9, C10, and C11 control the low-pass filter of the A3048BV1 amplifier. They must be equal to within ±2% to obtain a perfect response. When we buy a reel of ±5% capacitors, variation from one capacitor to the next in the reel has always been ±1%. We remove these three capacitors from all three boards and measure their values with two separate meters. We select new 2-nF capacitors from our own ±5% reel and measure each capacitor before loading onto the boards.
The original capacitors have values spanning 150 pF, which is ±3.8% of 2 nF. This explains why our frequency responses have degraded. Our replacements have values spanning 60 pF, or ±1.5%. We measure frequency response again with the new capacitors.
[17-JUL-24] Nathan reports. "We remove the logic chip from one of our A3048 circuits to measure the startup current from mostly just the ADC. We apply benchtop power to the circuit through a resistor in series. We use 10 ohms as well as 51 ohms and measure the potential across this resistor with a scope probe to view the spike in current consumption when the transmitter is switched on. We measure a peak current of about 20mA when measuring the 10 ohm resistor and a peak current of about 13mA when measuring the 51 ohm resistor."
The above result proves that it is the ADS8860 that draws 20 mA at startup, regardless of whether or not any other component is driving its inputs. The 100 kSPS ADS8866 is a drop-in replacement for the 1 MSPS ADS8860. We order 25 of these. Perhaps they consume less current on startup. They are also less expensive: $4.50 instead of $16 in quantity 100.
[18-JUL-24] Nathan tries the ADS8866 and finds the startup current is identical to that of the ADS8860 with a 51-Ω series resistor, see here. The area under the current trace is roughly 20 squares, or 20 × 4 mA × 0.5 ms = 40 μC. We recall our original obvservations of the power-up failure of the ADS8860 when we developed the A3047. When the power supplies rise to slowly, the ADC latches into a state where it consumes 2 mA and failes to digitize correctly. Because of this, we removed the 1-kΩ resistor between VD and VA that existed in the older A3028 circuit. When we turn on the transmitter, we connect C1 = C2 = 10 μF to the battery. One charges to 3 V, the other to 1.8 V, for a total of 50 μC.
[19-JUL-24] Nathan investigates. "We power an A3048 circuit with benchtop power through 10 ohms. This circuit has no logic chip and we are measuring the potential across the 10R resistor with a scope probe to detect startup current. We begin by removing C2 and observe that the transmitter still requires a peak current of 20vmA in order to start. We then load 100nF and 1uF in place of C2 and probe VA as well as VB at the same time as the startup current measurement. Finally, we remove the ADC entirely and leave 1uF loaded as C2. No change is observed after removing the ADC." He investigates further. "We take our A3048 circuit with removed logic and ADC and supply benchtop power through 10R in series. This time, we replace both C2 and C3 with 1uF. From top to bottom of the oscilloscope screen we observe VB, VA, and potential across the 10R resistor."
Startup charge is now 4 μC, down from 40 μC after reducing the capacitors C2 and C3 by a factor of ten. Hypothesis: The ADC makes no significant difference to the startup current. We suspect that the ADC requires a rapid, glitch-free power supply ramp-up in order to initialize correctly. In our original A3047 with 1-kΩ resistor in series with VA, the ramp-up was not fast enough. We eliminated the 1-kΩ, and now the ADC starts up correctly every time, but the current rushing into the circuit when it turns on increased by a factor of three or four. We take out an old A3028 circuit and measure its power-up current consumption. We see an inrush of around 1.5 μC, which is consistent with 10 μF charging to 1.8 V.
We start with three bare A3048AV1 circuits. We connect external batteries, short X to C, place in our Faraday enclosure and measure noise. We have No187, 2048 SPS, 6 μV, No188 512 SPS, 4 μV, and No190, 256 SPS, 4 μV. We load C1 = 22 μF, C2 = C3 = 1 μF. Now see No187, 2048 SPS, 5 μV, No188 512 SPS, 4 μV, and No190, 256 SPS, 4 μV. We conclude that there is no advantage to loading 10-μF capacitors onto VA and VL.
We prepare the bill of materials for the A3048BV2. We drop C5, C9, C10, C11 to 1800 pF so that we can use available 1% capacitors. Bandwidth increases to 90 Hz, but attenuition at the Nyquist frequency will not be as great. We drop C2 and C3 to 1 μF. We leave C1, C4, C8 at 10 μF.
[23-JUL-24] We drain four Premium CR927 batteries with 27 kΩ, see Premium_CR927. Operating life is around 250 hrs, or 25 mAhr. Operating voltage is around 2.7 V dropping to 2.5 V at 25 mAhr.
[25-JUL-24] Nathan makes four A3048BV2. He reports. "We change the ADC, 2 decoupling capacitors, and 3 capacitors in the amplifier on 4 of our A3048 circuits. We then program them to transmit at 64, 128, 256, 512, 1024, and 2048 SPS, measuring their characteristics each time. We record current consumption, startup current, startup charge, frequency response, and switching noise. One important thing to note is that we attempted to recreate the latching behavior observed in our A3048 circuits with the current BOM. We apply 1.0V to VB and slowly increase the voltage to 3.0V. We know that this causes the A3048BV1 circuits to latch up and consume roughly 2mA in its on state. With this new BOM change however we observe the transmitter unlatch itself after increasing VB past 2.5V." You will find Nathan's table of measurements of peak current, startup charge, noise, and μA/SPS in ADS8866_Table.
The −3 dB cut-off is now at around 90 Hz. The −20 dB cut-off remains at 130 Hz. We will continue to specify 80-Hz cut-off, even though the actual cut-off is slightly higher.
[31-JUL-24] Three transmitters from a batch of fifteen today failed sweep response with gain too high at 70 Hz. We resolve to replace C9, C10, C11 on all our BV1 boards. While we are at it, we will also replace C2 and C3 with 1 μF.
[04-SEP-24] Exerpt from an e-mail to two of our customers summarizing our diagnosis of five failures out of fifty transmitters implanted. Their experiment involves implanting, waiting a week, running for two weeks, waiting another two weeks, and running for two further weeks. The transmitter spends one two-week period turned off while implanted, and this appears to have increased the probability of failure from an overall average of about 2% to a crippling 10%.
"S234.117 turned on when we took it out of the package. It transmitted all zeroes. We know what causes the all-zero transmission: the analog to digital converter (ADC) has failed to initialize correctly. We remove the battery. It's voltage is 2.95 V, which is fine. Current consumption of the circuit is 45 uA, which is correct. When we connect the battery to the circuit, and help the battery start the circuit with an external power supply, the battery runs the circuit just fine. But the battery cannot power up the circuit correctly on its own.
"We have observed this same problem in two other transmitters you sent back to us. Now that we have started turning transmitters off and on in our accelerated aging tests, we have observed the problem in several transmitters that have been poaching for four or five weeks at 60C. We are calling this failure "corrosion-induced latch-up". The problem arises from a combination of three phenomena: temporary corrosion shorts in capacitors, a weakness in our circuit design that causes it to draw 20 mA when it powers up, and a vulnerability in our ADC that causes it to latch up if the power supplies are asserted too slowly. The first two phenomena have existed for as long as we have been making transmitters. They did not cause severe failures. In May 2023, however, we switched ADC because we could no longer buy the older ADC. The new ADC has the latch-up problem.
"We have a solution to this problem that we can implement immediately: we have made a simple change to our circuit that reduces the start-up current by a factor of ten, so that the battery has no trouble asserting the power supplies quickly, thus avoiding the latch-up in the ADC. Your most recent batch of transmitters includes this fix. In the long run, we have found a variant of this same ADC, studied the variant, and found that it does not have the same latch-up vulnerability. We are going to switch to the variant in future circuits, and this problem should go away completely.
"The interesting thing about the latch-up problem is this: it's much more likely to cause problems if you turn off the transmitter during your experiment, because as soon as you try to turn on the transmitter, you have to burn out any and all corrosion shorts that exist, and bring up the battery voltage fast enough to stop the ADC latching up, all at the same time. The longer you leave them implanted without running, the more likely the problem will occur. Thus it was you who had to suffer this problem the most, because you were turning of the transmitter for weeks 3-4, then turning it on again.
"The other source of failure in your transmitters was, we believe, flux residue beneath the ADC, which caused the growth of a metallic tendrils that drained the battery. We believe we will be able to avoid this problem with a more detailed cleaning protocol, where we specifically blast hot water under the ADC package to clean it before encapsulation.
[27-NOV-24] We have roughly 40 BV1 on the shelf. We will modify these to BV2 before transmitter construction. This modification loads 1.8 nF 1% for C9, C10, and C11, and loads 1.0 μF for C2 and C3. For future assemblies we define BV3 as BV2 with ADS8866 loaded for U7. The ADS8866 is the 100 kSPS version of the 1-MSPS ADS8860. The ADS8866 has one great advantage over the ADS8860, as demonstrated by the following experiment. We connect VB = 1.4 V to a BV3 circuit. We use a magnet to turn it on. It consumes 360 μA. We increase VB and record current consumption. We do the same for the BV2 circuit with ADS8860. We plot the results below.
The latch-up of the ADS8866 resolves once VB reaches 2.45 V. That of the ADS8860 fails to resolve even when VB reaches the absolute maximum rated value of 4.0 V. Given that our transmitter end of life is marked by battery voltage droppint below 2.5 V, the ADS8866 latch-up will never cause premature current drain.
We further modify our BV3 circuit, replacing C8 with 1.0 kΩ to make a DC-coupled amplifier with gain ×100. With all inputs open-circuit, uur AC-coupled BV1 and BV2 have average values 39162.7 and 39371.7 counts. Our DC-coupled BV3 has average 30077.0 cnt. We have an input offset voltage of 9.2k ÷ 66k × 30 mV = 4.1 mV. When we look at X with 10-MΩ probe, we see 500-mVpp 33-kHz square wave. This signal is being coupled into X from our TP2, which is carrying CK. We connect X and C. Our DC-coupled input rises to 36072 cnt, while the AC-coupled inputs remain constant, suggesting input offset 140 μV, which is consistent with the @OPA2369 ±250 μV specficiation.
We create P3048A05 firmware in which we remove CK from TP2 and replace with ECK. Average value of X in our DC-coupled BV3 is now 36840 cnt when input is open-circuit, 37528 when shorted. We add a linear shift register to the A05 firmware. We can examine the scatter easily by triggering on ECK and observing TP1, which carries FHI. We record forty-five scatter delays and plot.
In our current firmware, ECK occurs two CK period before the end of each sample interval. The ACTIVE period begins anything from 0 to 15 CK periods after the start of the next sample interval, or 2 to 17 periods after ECK. We find the average delay is 9.3 periods and the standard deviation is 4.5 periods, which is consistent with a random box distribution 16 periods wide. We see no sign of a repeating pattern of period 15 or smaller, which would be a sign of failure in the randomization of the LSR.
Open-circuit noise on all three of our test transmittters is 13 cnt rms, or 6 μV. When we short the inputs with a jumpber on the programming extension we see 5 counts rms, or 2.3 μVrms. The spectrum of open and short-circuit noise is white. The short-circuit noise is consistent with what we have been seeing for encapsulated A3048S2s in water. We see no difference between the noise in the BV3 with ADS8866 and the B1/B2 with ADS8860. We restore 10 μF for C8 in our BV3.
[29-NOV-24] We have one each of BV1, BV2, and BV3. We program with P4048A05, increasing sample rate from 64 to 2048 SPS. The BV3 is latched up consuming 500 μA almost every time we disconnect the programming cable from the circuit after programming. We turn on and off with a magnet and it recovers. The BV2 latches up only once, when we disconnect the cable, and consumes 2 mA. The BV1 latches up once as well, consuming 1.9 mA.
We release P3048A05 for production.
