The charge injection limit is partly an instrumentation artifact

What the work claims

Gholston, Sanchez, Bigger, Averbeck, Rieth, Ludwig, and Trevathan claim that two routine properties of the measurement setup, the input impedance of the recording instrument and the leakage current of the stimulator, have been systematically distorting published charge injection limits for platinum electrodes, and that thin-film microelectrode arrays are the most exposed. This is a primary experimental result plus a literature survey, not a grant or a perspective. The charge injection limit, the charge density at which interfacial polarization reaches the platinum water window of minus 0.6 V cathodic or plus 0.8 V anodic versus Ag|AgCl, is the number the whole field uses to set safe stimulation, and the paper demonstrates that the number is contingent on the front end used to measure it.¹

The design is deliberately instrument-centric. Three sizes of hand-polished 90-10 platinum-iridium disk electrodes, 1 mm, 0.250 mm, and 0.127 mm diameter, were characterized by electrochemical impedance spectroscopy and then measured at four effective input impedances: 1 megohm, 10 megohm, 10 gigaohm, and 10 teraohm, using a multimeter, its internal ranges, and the guard port of a current source to reach the highest values. Voltage transients used cathodic-leading, symmetric, nominally charge-balanced biphasic pulses, 200 microseconds per phase with a 50 microsecond interphase gap, at 30 Hz for 750 pulses, with 20 logarithmically spaced charge densities from 2 to 600 microcoulombs per square centimeter delivered by a current source chosen for its greater than 10 teraohm output impedance.¹

How it works

The physics is a voltage divider, which is what makes the result so uncomfortable. At DC a Randles-modeled electrode is purely resistive, the sum of charge transfer resistance and solution resistance, and that DC resistance is usually the highest impedance in the circuit. Against a measurement front end of input impedance Zin, the apparent electrode potential is attenuated by the ratio Rdc over Zin. A standard oscilloscope probe presents 1 to 10 megohms; an electrochemical potentiostat presents more than 10 teraohms. The paper's first result: at 1 megohm, apparent open circuit potentials for the two smaller electrodes collapse to 3.87 plus or minus 0.715 and 4.37 plus or minus 0.985 millivolts, against literature values of 250 to 300 millivolts for platinum in PBS, and even at 10 megohm they reach only 36.6 and 32.7 millivolts. Only at 10 gigaohm and above do all three sizes approach the expected range. The fitted charge transfer resistances explain why: 8.70 and 6.98 megaohm for the 0.127 and 0.250 mm electrodes against 0.241 megaohm for the 1 mm electrode, so probe loading scales exactly with the electrode size shrink that miniaturization demands.¹

Translated into safety numbers, the error is not cosmetic. The interpulse potential, used as the DC surrogate during pulse trains, converges only at high input impedance, settling between 1.1 and 1.3 V. Measured at 1 megohm without DC mitigation, the smallest electrodes appeared to tolerate an apparent charge injection limit of 250 plus or minus 15.6 microcoulombs per square centimeter; measured properly at 10 gigaohm the same configuration gave 4.71 plus or minus 0.894. An oscilloscope-front-end measurement overstates the safe limit for a small electrode by more than a factor of 50, because it cannot see the DC offset accumulating between pulses and therefore misplaces the electrode's operating point relative to the water window.¹

The second half of the mechanism is the stimulator itself. An inline DC blocking capacitor does not make a constant-current stimulator charge-balanced: stimulator leakage current keeps charging the capacitor until, around pulse 425 in the demonstrated case, the system hits the 20 V compliance limit, one phase of the biphasic waveform clips, and the pulse silently becomes charge-imbalanced, with the interpulse potential settling near 0.6 V for reasons unrelated to the electrode's electrochemistry. The fix tested is capacitive coupling plus shorting the working and counter electrodes between every pulse, built from 1 microfarad low-leakage capacitors and a high off-impedance analog switch. That combination reduced the average interpulse DC offset by 88 percent or more across charge densities. With leakage properly mitigated and impedance adequate, the apparent platinum limit for the smallest electrodes inverted its trend and rose to 359 plus or minus 13.8 microcoulombs per square centimeter at 10 gigaohm, above the historical 50 to 150 range derived from Robblee's 1990 protocol.¹

The survey of practice makes the stakes concrete. Of 77 studies meeting the inclusion criteria for reporting charge injection limits from voltage transients, 73 used an oscilloscope, only 3 reported the probe's input impedance, 72 reported the stimulator but only 23 reported any DC mitigation strategy, and 22 described a custom stimulator without specifications. Reported platinum limits across the literature span 150 to 5570 microcoulombs per square centimeter, and TiN spans 24 to 23000, a spread the authors attribute in significant part to this unreported instrumentation variability.¹

Where a skeptic should push

The most load-bearing assumption is that extrapolating from millimeter-and-submillimeter disk electrodes to true thin-film microelectrodes is monotone. The smallest electrode tested, at 1.27e-4 square centimeters, is still well above the 2e-5 square centimeter area the authors themselves cite as the microelectrode boundary, and the direction of the probe-loading error grows with Rdc, which grows as electrodes shrink. So the 50-fold apparent-limit inflation at 1 megohm is a lower bound on what a badly instrumented TFMEA measurement would show, not an estimate of it. The authors are commendably explicit on this and on a second caveat: their isolated single electrodes are a best case, free of the parasitic current paths that a dense array adds between neighboring contacts. The 10 gigaohm sufficiency threshold they establish should therefore be read as necessary, not sufficient, for array work.¹

The other push is on the water window itself as a safety framework. The paper uses the classical minus 0.6 V and plus 0.8 V platinum limits and acknowledges that corrosion chemistry, platinum chloride formation at roughly 0.5 V for instance, proceeds inside that window, so a limit measured against electrolysis onset is already permissive relative to material dissolution. The upward revision of platinum limits under proper instrumentation is real within that framework, but it is not a license, and the authors do not claim it is. Finally, the conflict-of-interest disclosure is dense: senior authors hold equity and advisory roles across neuromodulation companies and one is a principal investigator on industry contracts including an electrode manufacturer. That does not impeach the bench measurements, which are instrument physics, but it is worth stating because the reporting-framework recommendations interact with commercial test practice.¹

What a biased Qinj means for array stimulators

For microelectrode array hardware and the acquisition chain around it, this paper is about the difference between a measured number and a datasheet number. Closed-loop electrophysiology on living tissue increasingly means stimulating as well as recording, on the same small electrodes, and the safety case for those stimulation amplitudes usually traces back to an in vitro voltage-transient Qinj measured exactly the way this paper shows is broken: scope-class front end, unreported probe impedance, stimulator leakage unmitigated, 750 pulses in to a compliance rail nobody noticed. The uncomfortable corollary is that some published functional electrophysiology on arrays may carry a slow DC electric field at the tissue interface, from a stimulator that quietly stopped being charge-balanced around pulse 425, perturbing the very excitability being measured. Reproducibility failures across studies using thin-film arrays have an instrumentation suspect that no amount of spike-sorting refinement will find.¹

The non-obvious implication for array design is that DC tracking has to be an on-chip, per-channel function, not a benchtop afterthought. A wired MEA front end that saturates or loads at DC cannot distinguish electrode drift from amplifier offset; the paper's voltage-divider arithmetic applies to any amplifier whose input impedance is not orders of magnitude above the electrode's DC resistance, which for small, high-impedance contacts in saline is a real constraint at array scale, where input devices are sized for noise and bandwidth, not teraohm DC operation. The mitigation strategies that work at the bench, large blocking capacitors and inter-pulse shorting, are explicitly called out by the authors as impractical to scale to multichannel high-density implants, which pushes the problem into the ASIC: active charge balancing, per-electrode potential monitoring, and leakage current specification as a first-class design parameter rather than an unmeasured parasitic. A stimulator ASIC that does not state its output-stage leakage in the datasheet is, after this paper, simply underspecified.

The opportunity is headroom. If platinum's properly measured limit under honest instrumentation is genuinely above the 50 to 150 microcoulombs per square centimeter that three decades of equipment-confounded measurements entrenched, then stimulation designers have been leaving dynamic range on the table, and miniaturized electrodes, which need higher charge densities precisely because they are small, benefit most. The threat is symmetric and larger: every legacy safety margin, comparative materials claim, and regulatory submission built on unreconciled historical Qinj inherits the confound, and the correction direction is not uniform, it pushes down for leaky, low-impedance-measurement setups and up for clean ones. A field that has been comparing electrodes across studies that used unknowingly different instruments has been comparing instruments, not electrodes. The reporting table this paper issues, input impedance, stimulator leakage, output impedance, mitigation strategy, and the electrode's own Randles parameters, is the minimum disclosure an array instrumentation paper or datasheet should now be held to.¹

The bottom line

Established, in vitro on platinum-iridium: apparent open-circuit potential and interpulse DC offset are attenuated by measurement input impedance in quantitative agreement with a Randles voltage-divider model; at 1 megohm the smallest electrodes' apparent charge injection limit is inflated more than 50-fold relative to a 10 gigaohm measurement; capacitive coupling alone permits silent waveform clipping at stimulator compliance after roughly 400 pulses; adding inter-pulse electrode shorting cuts the residual DC offset by 88 percent or more; and with both confounds controlled, platinum limits rise above the historical 50 to 150 range. Not established: the limits for true thin-film microelectrode geometries, behavior inside dense arrays with inter-contact parasitics, corrosion behavior inside the water window, or in vivo confirmation, where protein adsorption and the foreign body response move every parameter. What would confirm the framework is a repeat on fabricated TFMEAs in array packaging, showing whether 10 gigaohm-class measurement and active charge balancing hold. What would break it is finding that in vivo DC accumulation dwarfs the in vitro effects, in which case the instrumentation correction, while true, would be the smaller term in the safety equation.