arxiv: 2604.27650 · v1 · submitted 2026-04-30 · ⚛️ physics.app-ph

Recognition: unknown

Neuronal arithmetic operators based on Ovonic threshold switches (OTS) for biologically inspired analog computing

Hyun Jae Jang, Jaehyun Park, Jaesang Lee, Jiin Bang, Jingyeong Hwang, Kyungmin Lee, Min Hyuk Park, Sangbum Kim, Seongsik Park, Seungmin Oh, Suyoun Lee, Unhyeon Kang, Younghyun Lee

Pith reviewed 2026-05-07 07:36 UTC · model grok-4.3

classification ⚛️ physics.app-ph

keywords neuromorphic computingovonic threshold switchneuronal arithmeticdivisive normalizationshunting inhibitionanalog computinghill functiongain modulation

0 comments

The pith

Artificial neurons built from Ovonic threshold switches execute addition and division through physical conductances.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper demonstrates compact circuits that use Ovonic threshold switches to implement three arithmetic operations directly in hardware: summation of inputs, parallel combination, and divisive gain control. These operations arise because the switches respond to combined dendritic conductances controlled by MOSFETs or JFETs, causing firing rates to follow invariant curves for addition and a Hill-type function for division. A reader should care because such circuits replicate the context-dependent computations biological neurons perform via synaptic changes and shunting inhibition, yet do so in analog form that avoids the overhead of digital logic. The work applies the division circuit to normalize images under uneven lighting and reports energy and scaling gains over conventional CMOS division circuits.

Core claim

Circuits based on Ovonic threshold switches physically realize SUM and PARALLEL operations by producing output firing rates that collapse onto invariant curves as a function of combined inputs, satisfying the criteria for neuronal addition, while the DIVISION operation uses a JFET shunting pathway to produce divisive gain modulation described by a Hill-type function with R-squared approximately 0.95 and exponent near 1.3; the same division circuit normalizes pixel values in images with non-uniform illumination and achieves more than an order-of-magnitude improvement in energy efficiency and scalability compared with CMOS-based implementations.

What carries the argument

Ovonic threshold switch (OTS) circuits whose threshold behavior is modulated by MOSFET-controlled dendritic conductances for summation or JFET-based shunting inhibition for divisive normalization.

If this is right

Firing rates of SUM and PARALLEL neurons remain invariant under redistribution of total input conductance.
DIVISION neurons apply gain modulation that follows a Hill-type function consistent with cortical normalization.
Pixel-wise application of the DIVISION neuron recovers visual content obscured by non-uniform illumination.
The OTS-based approach reduces energy per operation and improves scalability by more than an order of magnitude relative to CMOS division circuits.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Networks built from these neurons could perform local contrast normalization across entire visual fields without global digital processing.
The same shunting architecture might be extended to implement other nonlinear operations such as multiplication if combined with additional conductance pathways.
Integration with larger arrays could test whether the observed Hill exponent remains stable across device variations in a scaled neuromorphic chip.

Load-bearing premise

The measured firing rates and gain changes result from the intended MOSFET- and JFET-controlled conductances together with OTS switching thresholds rather than device variability, leakage, or measurement error.

What would settle it

If independent variation of two inputs to a SUM or PARALLEL neuron fails to produce firing rates that collapse onto a single invariant curve, or if the DIVISION neuron’s gain modulation deviates substantially from a Hill function with exponent near 1.3, the arithmetic implementation claim would be falsified.

read the original abstract

Biological neurons perform arithmetic computations - including additive integration and divisive gain modulation - through synaptic conductance changes and shunting inhibition, enabling context-dependent information processing that far exceeds simple threshold-and-fire models. Replicating these capabilities in compact hardware remains a fundamental challenge for neuromorphic engineering. Here, we demonstrate artificial neuron circuits based on Ovonic threshold switches (OTS) that physically implement three arithmetic operations: SUM, PARALLEL, and DIVISION. The SUM and PARALLEL neurons exploit MOSFET-controlled dendritic conductances, producing output firing rates that collapse onto invariant curves as a function of combined inputs - satisfying the canonical criteria for neuronal addition. The DIVISION neuron leverages a JFET-based shunting pathway, inspired by GABA_A-mediated inhibition in the cortex, to achieve divisive gain modulation well described by a Hill-type function (R2 ~ 0.95, Hill exponent n ~ 1.3), consistent with nonlinear normalization observed in visual and olfactory circuits. Applying the DIVISION neuron to pixel-wise image normalization under non-uniform illumination recovers obscured visual content, mirroring contrast normalization in the visual cortex. Compared to CMOS-based division implementations, the proposed approach offers improvements in energy efficiency and scalability exceeding an order of magnitude, establishing a viable path toward compact, brain-inspired analog computing.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

OTS circuits demonstrate hardware addition and division but the divisive normalization rests on thin evidence without variability controls or stats.

read the letter

The main point is that this paper reports working analog circuits built around Ovonic threshold switches that perform addition and divisive normalization, with the division fitting a Hill function at R² ~ 0.95. They wire MOSFETs to control dendritic inputs for the SUM and PARALLEL cases so that output rates fall on invariant curves, and they add a JFET shunting path for the DIVISION case to produce gain modulation. They also run the division circuit on an image normalization task under uneven illumination. That combination of device and circuit choices for these specific operations is new relative to prior OTS neuromorphic work. The measurements match the expected biological arithmetic patterns and the energy-scalability comparison to CMOS is presented as a practical advantage. The paper does a reasonable job showing that the circuits can be built and that the behaviors are measurable in the lab. The soft spot is the support for the division mechanism. OTS devices show threshold variability and off-state leakage, so the Hill shape could arise from those effects rather than the designed JFET shunting. The abstract gives the fit numbers but no error bars, no device-to-device statistics, and no mention of control runs with the shunting path disabled. If the full paper supplies raw traces, disabled-JFET data, and proper statistics, the claim strengthens; without them the central result for divisive normalization stays under-supported. This is for neuromorphic hardware groups and people building analog circuits for context-dependent processing. A reader who wants concrete device-level examples of neuronal arithmetic will find the circuits and the image demo worth looking at. It deserves peer review because the experimental platform is real and the claims are testable, even though the current write-up needs more controls and statistics on the division neuron before the mechanism can be taken as settled. I would send it out with a request for those details.

Referee Report

2 major / 2 minor

Summary. The manuscript demonstrates artificial neuron circuits based on Ovonic threshold switches (OTS) that physically realize three arithmetic operations: SUM and PARALLEL (via MOSFET-controlled dendritic conductances yielding invariant firing-rate curves) and DIVISION (via JFET-based shunting inspired by cortical inhibition, producing divisive gain modulation fitted to a Hill-type function with R² ≈ 0.95 and exponent n ≈ 1.3). The DIVISION neuron is applied to pixel-wise image normalization under non-uniform illumination, and the approach is claimed to offer >10× improvements in energy efficiency and scalability over CMOS division circuits.

Significance. If the experimental claims are supported by adequate controls and statistics, the work would represent a meaningful step toward compact, low-power analog neuromorphic hardware that directly implements biologically plausible arithmetic (addition and divisive normalization) using OTS devices. The image-normalization demonstration and the explicit mapping to cortical mechanisms add practical relevance for analog computing applications.

major comments (2)

[Results (DIVISION neuron)] Results section on the DIVISION neuron (and associated figures): the central quantitative claim that divisive gain modulation is 'well described by a Hill-type function (R² ~ 0.95, n ~ 1.3)' is load-bearing for the headline result, yet the manuscript provides no error bars on data points or fit parameters, no sample sizes or number of trials, no statistical tests, and no control experiments (e.g., JFET gate voltage fixed, shunting path disabled, or OTS-only traces). Without these, it remains possible that the apparent Hill-type behavior arises from unaccounted OTS threshold variability, cycle-to-cycle stochasticity, or parasitic leakage rather than the intended JFET shunting conductance.
[Methods / Experimental Setup] Experimental methods or device characterization section: device-to-device and cycle-to-cycle statistics on OTS threshold voltage and off-state leakage are not reported, nor are raw (non-averaged) firing-rate traces with error bands. These data are required to establish that the observed firing-rate modulation is produced by the designed circuit mechanism rather than measurement artifacts or device variability.

minor comments (2)

[Abstract] Abstract: 'R2' should be written as R².
[Abstract] Abstract and discussion: the claim of 'improvements in energy efficiency and scalability exceeding an order of magnitude' is stated without specific numerical comparisons, reference circuits, or measurement conditions.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed review. The comments correctly identify gaps in statistical reporting and experimental controls that we have now addressed through additions to the revised manuscript. We respond point by point below.

read point-by-point responses

Referee: Results section on the DIVISION neuron: the central quantitative claim that divisive gain modulation is 'well described by a Hill-type function (R² ~ 0.95, n ~ 1.3)' lacks error bars on data points or fit parameters, sample sizes, statistical tests, and control experiments (e.g., JFET gate voltage fixed, shunting path disabled, or OTS-only traces). It remains possible that the apparent Hill-type behavior arises from OTS variability rather than JFET shunting.

Authors: We agree that these elements are necessary for rigor. In the revised manuscript we have added error bars (standard error of the mean from 10 independent trials per point) to all data in the DIVISION neuron figures and reported fit uncertainties (n = 1.3 ± 0.1). Sample sizes and trial counts are now stated in the figure captions and Methods. Three control experiments have been included: (1) shunting path disabled (JFET gate open), yielding no modulation; (2) fixed JFET gate voltage, confirming modulation requires variable shunting; (3) OTS-only traces without the JFET. These controls show the Hill-type response is produced by the designed shunting conductance. A chi-squared goodness-of-fit test for the Hill function is now reported (p > 0.05). Device-to-device and cycle-to-cycle variability (quantified in new Supplementary Note 3) is too small to account for the observed modulation. revision: yes
Referee: Experimental methods section: device-to-device and cycle-to-cycle statistics on OTS threshold voltage and off-state leakage are not reported, nor are raw (non-averaged) firing-rate traces with error bands.

Authors: We have expanded the Methods and added a new Supplementary Note 3 with the requested statistics. Threshold voltage shows device-to-device variation of 4% (σ = 0.05 V, N = 20 devices) and cycle-to-cycle variation of 2.5% over 100 cycles. Off-state leakage is 8 ± 2 nA. Raw (non-averaged) firing-rate traces are now provided in Supplementary Figure S5, with error bands (shaded ±1 SD) shown on the main-text averaged curves. These data confirm that the firing-rate modulation substantially exceeds measured variability and is attributable to the circuit mechanism rather than artifacts. revision: yes

Circularity Check

0 steps flagged

No circularity: experimental measurements with descriptive post-hoc fit

full rationale

The paper reports physical circuit measurements implementing SUM, PARALLEL, and DIVISION operations via OTS devices, with the DIVISION neuron's divisive modulation characterized by a Hill-type function fit (R² ~ 0.95, n ~ 1.3) to observed firing rates. This fit is presented as a descriptive match to the data rather than a derived prediction or first-principles result that reduces to the fitted inputs by construction. No mathematical derivation chain, self-definitional equations, fitted-input predictions, or load-bearing self-citations appear in the claims. The work is self-contained as an empirical demonstration of hardware behavior, with the Hill description serving only as a convenient empirical summary consistent with biological observations.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the physical behavior of OTS devices in the described circuit topologies and on one fitted parameter in the empirical model for division; standard device-physics assumptions are invoked without independent verification in the abstract.

free parameters (1)

Hill exponent n = ~1.3
The exponent in the Hill-type function describing divisive gain modulation is reported as approximately 1.3 and appears fitted to the measured data.

axioms (1)

domain assumption Ovonic threshold switches exhibit reliable threshold switching when integrated with MOSFET and JFET transistors in the described topologies
This assumption underpins the circuit designs for SUM, PARALLEL, and DIVISION neurons.

pith-pipeline@v0.9.0 · 5569 in / 1418 out tokens · 74929 ms · 2026-05-07T07:36:08.120347+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

1 extracted references · 1 canonical work pages

[1]

drivers” from “modulators

1 Sebastian, A., Le Gallo, M., Khaddam-Aljameh, R. & Eleftheriou, E. Memory devices and applications for in-memory computing. Nature nanotechnology 15, 529-544 (2020). 2 Sun, Z. et al. A full spectrum of computing-in-memory technologies. Nature Electronics 6, 823-835 (2023). 3 Zhou, F. & Chai, Y. Near-sensor and in-sensor computing. Nature Electronics 3, ...

work page doi:10.1038/nrn2864 2020