Pith. sign in

REVIEW 3 major objections 5 minor 73 references

Machine Intelligence on Wireless Edge Networks

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The radio receive chain already in billions of edge devices can run neural inference in analog, the paper argues, with weights broadcast as RF waveforms and multiply–accumulate done by the existing mixer and filter stages.

desk verdict The diode-mixer energy-window analysis is real and checkable, but the 'no extra hardware' claim needs a demonstrated client-side activation injection path and a channel model before it can be believed. read the letter →

arxiv 2506.12210 v2 pith:STONISQK submitted 2025-06-13 cs.ET cs.LG

Sri Krishna Vadlamani , Kfir Sulimany , Zhihui Gao , Tingjun Chen , Dirk Englund This is my paper · ORCID
classification cs.ETcs.LG
keywords in-physicscomputationwirelessedgeinferenceRFreceivechaindioderingmixerfrequency-multiplexedencodinganaloginnerproductenergy-accuracytrade-offdisaggregateddeeplearning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper proposes turning the radio receive chain of edge devices into the compute engine for neural-network inference. A server broadcasts the model's weights as radio-frequency waveforms; the client encodes its own sensor data as a second waveform, and the mixer and filter stages already present in the receiver perform the multiply–accumulate operations directly in analog. The device therefore neither stores the model nor repeatedly converts between analog and digital domains, and the broadcast cost is spread across every client. The paper argues that thermal noise and diode-mixer nonlinearity leave an optimal energy window in which analog inner products stay accurate, and that training through a differentiable model of the RF chain preserves classification accuracy inside that window. On the MNIST benchmark the scheme reaches about 95 percent accuracy at 100 picojoules of client input energy, which the paper says lies within the typical energy budgets of standard front-end receivers.

What carries the argument

The load-bearing object is the diode ring mixer, a passive four-diode circuit already used for frequency conversion in RF receivers, whose output is the nonlinear function of its two inputs $y(t) = \frac{w(t)}{2} + \frac{V_T}{2}\ln\left(\frac{e^{x(t)/V_T}+e^{-w(t)/V_T}}{e^{x(t)/V_T}+e^{w(t)/V_T}}\right)$ with $V_T = kT/e$ the thermal voltage. At small amplitudes this expression reduces to the exact product $\frac{w(t)x(t)}{4V_T}$, so the mixer behaves as an analog multiplier; at larger amplitudes the higher-order terms take over and the output departs from multiplication, which sets the high-energy edge of the accuracy window. Around this sits the frequency-comb encoding: weights occupy finely spaced comb lines, activations coarsely spaced ones, so that each output comb line selected by the band-pass filter carries one component of the matrix–vector product. Johnson–Nyquist noise at the two input ports and the output sets the low-energy edge, and the accuracy metric $\mathrm{ENOB} = \frac{1}{2}\log_2(1+\mathrm{SNR})$, derived from the mutual information between the noisy analog result and the exact digital inner product, locates the optimal operating energy. The hardware-tailored pipeline treats the same mixer transfer function as a differentiable layer, so training adapts the network to the hardware's nonlinearity instead of assuming an ideal multiplier.

What would settle it

Take a real diode ring mixer, drive it with two frequency-comb waveforms generated exactly as in Sec. 2.1.2, read off the appropriate output comb line, and compare the measured ENOB-versus-energy curve with Fig. 2; the central claim fails if the measured peak accuracy, or the input power at which it occurs, disagrees with the predicted window. The deployment claim fails if a broadcast weight comb passed over an indoor multipath channel no longer yields the expected matrix–vector product after mixing, which a direct over-the-air test would reveal.

Watch

Extended reading notes

Core claim

The paper's central claim is that the same RF chain used for communication in edge devices can also carry out machine-learning inference with accuracy close to digital implementations, when operated within the typical energy constraints of standard front-end receivers. The mechanism is frequency-multiplexed encoding: the server packs a weight matrix into the amplitudes of finely spaced frequency-comb lines of a broadcast waveform $w(t)$, the client packs its activation vector into a coarsely spaced comb $x(t)$, and the diode ring mixer—a passive four-diode circuit that multiplies two RF waveforms approximately—produces output comb lines whose amplitudes, after band-pass filtering, are exactly the components of the matrix–vector product $Wx$. The paper then shows that Johnson–Nyquist thermal noise at low power and the mixer's nonlinear transfer function at high power bound the accuracy of these analog inner products, producing an energy–accuracy curve with a peak; the effective number of bits (ENOB), defined as $\frac{1}{2}\log_2(1+\mathrm{SNR})$ and derived from the mutual information between the analog and digital inner products, quantifies that peak. Two end-to-end pipelines are offered: a digital-tailored one that folds the Fourier transforms into the broadcast weights via IQ modulation so the client never computes an FFT, and a hardware-tailored one that encodes in the time domain and trains through a differentiable model of mixing, filtering, and an analog layer-norm, eliminating analog-to-digital conversion between layers. Circuit-informed simulations, which the paper reports as consistent with a companion experiment, reach roughly 95 percent accuracy on MNIST at 100 picojoules of client input energy; the paper calls this family of architectures Machine Intelligence on Wireless Edge Networks (MIWEN).

Load-bearing premise

The load-bearing premise is that the client can inject its own activation waveform into the receive chain's mixer port and that the broadcast weight waveform arrives at that port with its comb-line amplitudes intact; stock receivers provide no such injection point, and realistic wireless channels distort broadcast signals, yet the simulations treat both as ideal.

Editorial extensions

If this is right

  • Client devices no longer need to store the model: weights stream from the server as RF waveforms, so model capacity is decoupled from device memory and the broadcast cost is amortized across every client.
  • In the hardware-tailored variant, inference runs without dedicated accelerator hardware and without analog-to-digital conversion between layers, shifting the energy cost of edge AI to RF front-end operations the device already performs for communication.
  • Accuracy is bounded by an energy window rather than by model size: near-digital accuracy around 95 percent on MNIST holds near 100 pJ of client input energy, and pushing energy higher degrades accuracy as the diode mixer departs from ideal multiplication.
  • Training must be hardware-aware: networks trained through a differentiable model of the mixer, filter, and analog layer-norm recover accuracy that a model assuming an ideal multiplier would not achieve in this physics.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the broadcast weight signal is a multi-carrier comb much like OFDM subcarriers, a natural testable extension is whether standard per-subcarrier channel estimation and equalization could compensate multipath distortion of the weights so the scheme survives realistic links; the paper leaves this open since its simulations assume undistorted arrival.
  • The broadcast architecture implies all clients receive identical weights, so as described it serves shared server-owned models but not per-device personalization; hybrid designs would need on-device adaptation that the paper does not specify.
  • The ENOB-as-mutual-information metric is portable: any analog inner-product engine, such as in-memory crossbar arrays or photonic dot-product cores, could be scored the same way against its digital ideal, and a similar noise-versus-nonlinearity window should appear.
  • A testable prediction of the energy-window analysis is that replacing the passive diode ring with a more linear active mixer should push the peak accuracy to lower energies; the paper does not make this prediction explicitly.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes MIWEN, an architecture in which a base station broadcasts neural-network weights as frequency-multiplexed RF waveforms and an edge device performs matrix-vector products by multiplying the received weight waveform with a locally generated activation waveform in the diode-ring mixer of its existing receive chain. The manuscript derives the diode-ring mixer transfer function, analyzes thermal noise and nonlinearity to define an effective-number-of-bits (ENOB) metric, and identifies an optimal energy window for analog inner products. It then presents two training pipelines: a digital-tailored variant that uses IQ modulators to merge Fourier transforms into the weights, and a hardware-tailored variant that trains through a differentiable model of mixer, LayerNorm, and filter stages. Numerical simulations on MNIST report near-digital accuracy at around 100 pJ activation energy before performance degrades at higher energies. The central claim is that this computation can happen in unmodified commodity RF front ends without additional hardware or repeated signal conversions.

Significance. If the architecture performed as claimed, it would be a meaningful contribution: it would remove local model storage from edge inference, amortize weight transmission over many clients, and reuse RF components already present in mobile devices. The paper has real strengths: the code and data are publicly available; the ENOB benchmark is anchored to the ideal digital inner product rather than to its own output; the small-signal and exact diode-ring derivations in Methods 4.3 are explicit; and the hardware-tailored training-through-physics approach is a sensible way to compensate for analog nonidealities. The reported energy-accuracy trade-off is an internally consistent consequence of the modeled noise and nonlinearity. However, the headline claims outrun the evidence: the simulations assume an ideal wireless channel and an ideal client-side activation-waveform generator/injection path, and the energy accounting excludes major client-side components. The contribution is therefore conditional on modeling and validating those missing pieces.

major comments (3)
  1. [Sec. 2.1.2, Sec. 2.2.2, Sec. 2.3.3] The central claim that inference can be performed by 'existing mixer and filter stages' in unmodified commodity receivers is not supported by the simulations. Eqs. (1)-(3) define w(t) and x(t) as waveforms to be inserted directly into the mixer transfer function Eq. (5), and the custom PyTorch layers in Sec. 2.3.3 do the same; there is no model of how x(t) is generated on the client or how it is injected into the mixer's second port. In a standard receiver, that port is driven by a fixed local oscillator, not by an arbitrary activation comb. Generating x(t) with d frequency components requires a DAC/arbitrary-waveform generator and an IQ modulator or oscillator bank; the digital-tailored pipeline additionally requires IQ modulators/demodulators, switches, and time-bin nonlinearities (Sec. 2.3.1, Fig. 3), and the hardware-tailored pipeline requires op-amp LayerNorm and extra filters (Sec. 2.3.2, Fig. 4). These components are not 'existing mixer and filter stages,' and their energy is not included in Fig. 2 (whose caption excludes ADC energy) or in Fig. 5's total-energy curves (which, per Sec. 2.3.3, add only LayerNorm amplification energy). The abstract's claims of eliminating 'repeated signal conversions and extra hardware' and operating 'within the typical energy constraints of standard front-end receivers' are therefore not established. The claim should be weakened to a conditional statement, or the activation-generation/injection path should be modeled with its energy cost.
  2. [Secs. 2.1.2, 2.2.2, 2.3.3] The wireless channel is treated as ideal. Although Sec. 2.1.2 states that weights are streamed over a wireless channel, the simulations feed the transmitted waveform directly into Eq. (5) without path loss, multipath, fading, carrier-frequency offset, or synchronization error. The ENOB contours in Fig. 2 and the accuracy curves in Fig. 5 are computed from the mixer model and thermal noise only. Because the analog inner product is sensitive to the amplitudes and phases of the received comb lines, a realistic channel could shift the claimed optimal energy window or destroy the orthogonality of the encoding. The manuscript should either include a channel model, even a simple additive-noise plus path-loss model, or limit the stated claims to wired or back-to-back settings.
  3. [Methods 4.1, Eq. (8)] The ENOB formula is exact only for additive Gaussian noise whose variance is independent of the signal. In Sec. 4.1.2, the paper replaces the analog error with n' = f(w+n_w, x+n_x) - w*x + n, which for the diode-ring mixer of Eq. (5) is nonlinear in w and x and hence neither Gaussian nor signal-independent, yet the same I = (1/2) log2(1+SNR) expression is used without justification. This matters because Fig. 2 uses ENOB to define the claimed optimal energy window. The authors should test the approximation, for example by comparing Eq. (8) with a direct Monte Carlo estimate of the mutual information between w*x and the analog output, and report the discrepancy.
minor comments (5)
  1. [Methods 4.1.1] The text uses 'yoptical' in the conditional-entropy sentence; this should read 'yanalog'.
  2. [Methods 4.3.2 and Eq. (5)] The variable names are inconsistent: Eq. (5) uses w and x, while the derivation in Methods 4.3.2 uses VLO and VRF. Please state the mapping between these variables explicitly.
  3. [Fig. 5 caption and Sec. 2.3.3] The caption says 'total energy per inference,' but the text describes client-side input energy, and the total-energy markers include only activation plus LayerNorm amplification. Please reconcile the terminology and state clearly which client-side components are included and which are excluded.
  4. [Fig. 2 caption] The bottom axes are labeled 'power (W)' while the top axes are labeled 'energy per MAC (J)'; the integration time or bandwidth used to convert between power and energy per MAC is not stated.
  5. [References] Refs. [66] and [71] are prior works by the same group that supply the companion experiment and the frequency-encoding method, respectively; the manuscript should state explicitly which components of MIWEN are new here and which are established in those references, especially regarding the client-side activation injection path.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: ENOB and MNIST accuracy are externally benchmarked; self-citations are contextual, not load-bearing.

full rationale

The paper's central quantitative claims are externally anchored rather than equivalent to their inputs. The frequency-multiplexed matrix-vector identity (Eq. 4) is proved in-text from the Fourier comb definitions in Eqs. (1)-(3); the diode-ring mixer transfer function (Eq. 5) is derived from the diode characteristic in Methods 4.3; and the ENOB metric (Eq. 8) is benchmarked against the digital inner product w·x, with the signal defined as the squared ideal product and the noise as the physical deviation. The MNIST results in Fig. 5 are obtained by training through a differentiable simulation of the mixer-filter chain and measuring classification accuracy against ground-truth labels, an external benchmark that does not presuppose the conclusion. The energy-accuracy trade-off follows from the circuit model and the definition of ENOB, not from the target accuracy values. The paper cites its own companion experiment [66] and prior optical frequency-encoding work [71], but the load-bearing equations and simulations are self-contained; these citations provide context and a consistency check rather than the derivation itself. The principal limitation is an unsupported hardware-feasibility assumption: the client-side activation waveform is assumed to be generated and injected into a stock receive-chain mixer port, with the associated DAC/IQ-modulator energy excluded and not modeled in the simulations. That is an engineering gap, not a circular derivation. No step in the derivation chain reduces to its own output by construction.

Assumptions & free parameters 0 free parameters · 7 assumptions · 0 invented entities

All simulation results rest on a small set of physical and architectural assumptions. The mixer transfer function in Eq. (5) is a domain model of a diode ring mixer; real mixers have imbalances and frequency responses. Noise is modeled only as Johnson-Nyquist 4kTRΔf, with shot noise ignored per Sec. 2.2.1. The frequency-domain extraction assumes ideal bandpass filtering. Critically, the wireless channel is not modeled at all in Sec. 2.3.3, so the received weight waveform is implicitly identical to the transmitted one. The architecture also assumes the client can create and route activation waveforms into the mixer without the transmit path or extra DAC/upconverter. The ENOB analysis invokes the central limit theorem and zero-mean independent inputs (Methods 4.1, 4.2). These are all stated or implicit assumptions, not fitted parameters.

assumptions (7)
  • domain assumption Diode ring mixer transfer function is Eq. (5): y = w/2 + VT/2 ln((e^{x/VT}+e^{-w/VT})/(e^{x/VT}+e^{w/VT})).
    All accuracy and ENOB simulations use this model as the physical mixer (Sec. 2.2.1, Methods 4.3); real mixers have imbalances and frequency-dependent responses not captured.
  • domain assumption Johnson-Nyquist noise with variance 4kTRΔf is the only noise source; shot noise is ignored.
    Stated in Sec. 2.2.1: 'The electronic shot noise in the system is ignored.' This simplifies the noise analysis.
  • domain assumption The wireless channel is ideal: the received weight waveform equals the transmitted waveform.
    Sec. 2.3.3 simulations feed waveforms directly into the mixer model without path loss, fading, or synchronization, despite the abstract claiming 'realistic wireless edge scenarios'.
  • domain assumption Ideal bandpass filtering separates the desired MVM comb line and blocks all spurious frequencies.
    Sec. 2.1.2: 'use a bandpass filter to pass all the components of W·x* while blocking all other spurious frequencies.' No filter non-idealities are modeled.
  • domain assumption The client can generate and inject local activation waveforms x(t) into receive-chain mixer ports without additional hardware.
    Protocols in Sec. 2.1.2 and Fig. 4 assume a controllable activation waveform at the mixer input; commodity receive chains do not expose such an input, and generating x(t) at RF requires a DAC and upconverter.
  • standard math Central Limit Theorem: w·x with independent zero-mean components is Gaussian.
    Methods 4.1.1 uses CLT to derive the Gaussian distribution of w·x and the ENOB formula.
  • domain assumption Weights and activations are independent zero-mean random variables with variances M_w^2, M_x^2.
    Methods 4.2.1 assumes unit normal sampling to compute SNR and ENOB, and Fig. 2 uses this distribution.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Machine Intelligence on Wireless Edge Networks." pith.science (2026). https://pith.science/paper/STONISQK

@misc{pith2026250612210,
  author       = {Pith},
  title        = {Pith review of: Machine Intelligence on Wireless Edge Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/STONISQK}},
  note         = {Machine review of arXiv:2506.12210}
}
read the original abstract

Machine intelligence on edge devices enables low-latency processing and improved privacy, but is often limited by the energy and delay of moving and converting data. Current systems frequently avoid local model storage by sending queries to a server, incurring uplink cost, network latency, and privacy risk. We present the opposite approach: broadcasting model weights to clients that perform inference locally using in-physics computation inside the radio receive chain. A base station transmits weights as radio frequency (RF) waveforms; the client encodes activations onto the waveform and computes the result using existing mixer and filter stages, RF components already present in billions of edge devices such as cellphones, eliminating repeated signal conversions and extra hardware. Analysis shows that thermal noise and nonlinearity create an optimal energy window for accurate analog inner products. Hardware-tailored training through a differentiable RF chain preserves accuracy within this regime. Circuit-informed simulations, consistent with a companion experiment, demonstrate reduced memory and conversion overhead while maintaining high accuracy in realistic wireless edge scenarios.

Figures

Figures reproduced from arXiv: 2506.12210 by the authors.

Figure 1
Figure 1. System illustration. The edge device utilizes the existing radio frequency (RF) analog chain of wireless transceivers to perform disaggregated deep learning. The weights, denoted as w, are transmitted as RF signals from a server (left), encoded in either the frequency or time domain. These signals are received by the edge device, where they are multiplied by the data signals using an RF diode ring mixer. The signal … view at source ↗
Figure 2
Figure 2. Effective number of bits (ENOB) for in￾ner product computation vs power consumption. ENOB in the output of the radio frequency inner prod￾uct computation w · x is plotted as a function of client and server energy per Multiply-Accumulate (MAC) for a diode ring mixer. Here “client energy per MAC” ac￾counts for the client’s activation waveform energy at a 50 Ω input; “server energy per MAC” accounts for the transmitted… view at source ↗
Figure 3
Figure 3. Digital-tailored machine learning computation pipeline. In our digital-tailored proposal, the client only makes use of IQ modulators and demodulators and completely eschews the digital FFT and IFFT operations. Spurious frequency produced by the multiplication process appear as spurious time bins after IQ demodulation and are temporally filtered out. The activations are passed through a time-bin nonlinearity before b… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Hardware-tailored machine learning computation pipeline. The server sends waveform w(t) that gets mixed with the local time-encoded activa￾tions waveform x(t) at the client’s side. The result is sent through a layernorm layer (implemented via opamps) to stabilize the a…
Figure 5
Figure 5. Figure 5: presents the variation in test accuracy as a function of client energy for two different network archi￾tectures (smaller: 49-32-16-10 and larger: 196-64-32-10) across two energy-accounting modes: activation energy and total energy. The energy thresholds are applied to …
Figure 6
Figure 6. Figure 6: Schematic of the passive diode ring mixer. The [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

73 extracted references · 66 canonical work pages

  1. [1]

    Efficient processing of deep neural networks: A tutorial and survey

    Vivienne Sze, Yu-Hsin Chen, Tien-Ju Yang, and Joel S Emer. Efficient processing of deep neural networks: A tutorial and survey. Proceedings of the IEEE, 105(12):2295–2329, 2017

    work page 2017

  2. [2]

    Computing’s energy problem (and what we can do about it)

    Mark Horowitz. Computing’s energy problem (and what we can do about it). In Proc. IEEE International Solid-State Circuits Conference (ISSCC), 2014

    work page 2014

  3. [3]

    In-datacenter per- formance analysis of a tensor processing unit

    Norman P Jouppi, Cliff Young, Nishant Patil, David Patter- son, Gaurav Agrawal, Raminder Bajwa, Sarah Bates, Suresh Bhatia, Nan Boden, Al Borchers, et al. In-datacenter per- formance analysis of a tensor processing unit. In Proc. IEEE/ACM International Symposium on Computer Archi- tecture (ISCA), 2017

    work page 2017

  4. [4]

    The emergence of edge computing

    Mahadev Satyanarayanan. The emergence of edge computing. Computer, 50(1):30–39, 2017. 12

    work page 2017

  5. [5]

    A method to estimate the energy consumption of deep neu- ral networks

    Tien-Ju Yang, Yu-Hsin Chen, Joel Emer, and Vivienne Sze. A method to estimate the energy consumption of deep neu- ral networks. In 2017 51st asilomar conference on signals, systems, and computers, pages 1916–1920. IEEE, 2017

    work page 2017

  6. [6]

    Split learning for health: Distributed deep learning without sharing raw patient data

    Praneeth Vepakomma, Otkrist Gupta, Tristan Swedish, and Ramesh Raskar. Split learning for health: Distributed deep learning without sharing raw patient data. arXiv preprint arXiv:1812.00564, 2018

    arXiv 2018

  7. [7]

    Multi-key privacy-preserving deep learning in cloud computing

    Ping Li, Jin Li, Zhengan Huang, Tong Li, Chong-Zhi Gao, Siu- Ming Yiu, and Kai Chen. Multi-key privacy-preserving deep learning in cloud computing. Future Generation Computer Systems, 74:76–85, 2017

    work page 2017

  8. [8]

    Quantum-secure multi- party deep learning

    Kfir Sulimany, Sri Krishna Vadlamani, Ryan Hamerly, Prahlad Iyengar, and Dirk Englund. Quantum-secure multi- party deep learning. arXiv preprint arXiv:2408.05629, 2024

    arXiv 2024

Show all 73 references
  1. [9]

    Wright, Peter L

    Ali Momeni, Babak Rahmani, Benjamin Scellier, Logan G. Wright, Peter L. McMahon, Clara C. Wanjura, Yuhang Li, Anas Skalli, Natalia G. Berloff, Tatsuhiro Onodera, Ilker Oguz, Francesco Morichetti, Philipp del Hougne, Manuel Le Gallo, Abu Sebastian, Azalia Mirhoseini, Cheng Zhan...

    2025

  2. [10]

    Backpropagation-free train- ing of deep physical neural networks

    Ali Momeni, Babak Rahmani, Matthieu Mall´ ejac, Philipp Del Hougne, and Romain Fleury. Backpropagation-free train- ing of deep physical neural networks. Science, 382(6676): 1297–1303, 2023

    2023

  3. [11]

    Deep physical neural networks trained with backpropagation

    Logan G Wright, Tatsuhiro Onodera, Martin M Stein, Tianyu Wang, Darren T Schachter, Zoey Hu, and Peter L McMahon. Deep physical neural networks trained with backpropagation. Nature, 601(7894):549–555, 2022

    2022

  4. [12]

    Inference in artificial intelligence with deep optics and photonics

    Gordon Wetzstein, Aydogan Ozcan, Sylvain Gigan, Shanhui Fan, Dirk Englund, Marin Soljaˇ ci´ c, Cornelia Denz, David AB Miller, and Demetri Psaltis. Inference in artificial intelligence with deep optics and photonics. Nature, 588(7836):39–47, 2020

    2020

  5. [13]

    Memory devices and applications for in-memory computing

    Abu Sebastian, Manuel Le Gallo, Riduan Khaddam-Aljameh, and Evangelos Eleftheriou. Memory devices and applications for in-memory computing. Nature Nanotechnology, 15(7):529– 544, 2020

    2020

  6. [14]

    Photonics for artificial intelligence and neuro- morphic computing

    Bhavin J Shastri, Alexander N Tait, Thomas Ferreira de Lima, Wolfram HP Pernice, Harish Bhaskaran, C David Wright, and Paul R Prucnal. Photonics for artificial intelligence and neuro- morphic computing. Nature Photonics, 15(2):102–114, 2021

    2021

  7. [15]

    Fully forward mode training for optical neural networks

    Zhiwei Xue, Tiankuang Zhou, Zhihao Xu, Shaoliang Yu, Qionghai Dai, and Lu Fang. Fully forward mode training for optical neural networks. Nature, 632(8024):280–286, 2024

    2024

  8. [16]

    Nonlinear processing with lin- ear optics

    Mustafa Yildirim, Niyazi Ulas Dinc, Ilker Oguz, Demetri Psaltis, and Christophe Moser. Nonlinear processing with lin- ear optics. Nature Photonics, 18(10):1076–1082, 2024

    2024

  9. [17]

    The physics of optical computing

    Peter L McMahon. The physics of optical computing. Nature Reviews Physics, 5(12):717–734, 2023

    2023

  10. [18]

    An on-chip photonic deep neural network for image classifica- tion

    Farshid Ashtiani, Alexander J Geers, and Firooz Aflatouni. An on-chip photonic deep neural network for image classifica- tion. Nature, 606(7914):501–506, 2022

    2022

  11. [19]

    Parallel convolutional processing using an integrated photonic tensor core

    Johannes Feldmann, Nathan Youngblood, Maxim Karpov, Helge Gehring, Xuan Li, Maik Stappers, Manuel Le Gallo, Xin Fu, Anton Lukashchuk, Arslan S Raja, et al. Parallel convolutional processing using an integrated photonic tensor core. Nature, 589(7840):52–58, 2021

    2021

  12. [20]

    Deep learning with coherent vcsel neural networks

    Zaijun Chen, Alexander Sludds, Ronald Davis III, Ian Chris- ten, Liane Bernstein, Lamia Ateshian, Tobias Heuser, Niels Heermeier, James A Lott, Stephan Reitzenstein, et al. Deep learning with coherent vcsel neural networks. Nature Photon- ics, 17(8):723–730, 2023

    2023

  13. [21]

    An optical neural network using less than 1 photon per multiplication

    Tianyu Wang, Shi-Yuan Ma, Logan G Wright, Tatsuhiro On- odera, Brian C Richard, and Peter L McMahon. An optical neural network using less than 1 photon per multiplication. Nature Communications, 13(1):123, 2022

    2022

  14. [22]

    11 tops photonic convolutional accelerator for optical neural networks

    Xingyuan Xu, Mengxi Tan, Bill Corcoran, Jiayang Wu, An- dreas Boes, Thach G Nguyen, Sai T Chu, Brent E Little, Damien G Hicks, Roberto Morandotti, et al. 11 tops photonic convolutional accelerator for optical neural networks. Nature, 589(7840):44–51, 2021

    2021

  15. [23]

    Programmable photonic circuits

    Wim Bogaerts, Daniel P´ erez, Jos´ e Capmany, David AB Miller, Joyce Poon, Dirk Englund, Francesco Morichetti, and Andrea Melloni. Programmable photonic circuits. Nature, 586 (7828):207–216, 2020

    2020

  16. [24]

    A three- terminal nanophotonic integrator for deep neural networks

    Saumil Bandyopadhyay, Kfir Sulimany, Alexander Sludds, Ryan Hamerly, Keren Bergman, and Dirk Englund. A three- terminal nanophotonic integrator for deep neural networks. In 2025 Optical Fiber Communications Conference and Exhibi- tion (OFC), pages 1–3. IEEE, 2025

    2025

  17. [25]

    Qamnet: Fast and efficient optical qam neural networks

    Marc Gong Bacvanski, Sri Krishna Vadlamani, Kfir Sulimany, and Dirk Robert Englund. Qamnet: Fast and efficient optical qam neural networks. arXiv preprint arXiv:2409.12305, 2024

    2024 arXiv

  18. [26]

    Attojoule optoelectronics for low-energy information processing and communications

    David AB Miller. Attojoule optoelectronics for low-energy information processing and communications. IEEE Journal of Lightwave Technology, 35(3):346–396, 2017

    2017

  19. [27]

    Analog optical computer for ai inference and com- binatorial optimization

    Natalia Berloff, Kirill Kalinin, Hitesh Ballani, and Francesca Parmigiani. Analog optical computer for ai inference and com- binatorial optimization. 2025

    2025

  20. [28]

    Large-scale photonic chiplet taichi empowers 160-tops/w artificial general intelligence

    Zhihao Xu, Tiankuang Zhou, Muzhou Ma, ChenChen Deng, Qionghai Dai, and Lu Fang. Large-scale photonic chiplet taichi empowers 160-tops/w artificial general intelligence. Sci- ence, 384(6692):202–209, 2024

    2024

  21. [29]

    All-optical ma- chine learning using diffractive deep neural networks

    Xing Lin, Yair Rivenson, Nezih T Yardimci, Muhammed Veli, Yi Luo, Mona Jarrahi, and Aydogan Ozcan. All-optical ma- chine learning using diffractive deep neural networks. Science, 361(6406):1004–1008, 2018

    2018

  22. [30]

    Isaac: A convolutional neural network accelerator with in-situ analog arithmetic in cross- bars

    Ali Shafiee, Anirban Nag, Naveen Muralimanohar, Rajeev Balasubramonian, John Paul Strachan, Miao Hu, R Stanley Williams, and Vivek Srikumar. Isaac: A convolutional neural network accelerator with in-situ analog arithmetic in cross- bars. ACM SIGARCH Computer Architecture News,...

    2016

  23. [31]

    Prime: A novel processing-in-memory architecture for neural network compu- tation in reram-based main memory

    Ping Chi, Shuangchen Li, Cong Xu, Tao Zhang, Jishen Zhao, Yongpan Liu, Yu Wang, and Yuan Xie. Prime: A novel processing-in-memory architecture for neural network compu- tation in reram-based main memory. ACM SIGARCH Com- puter Architecture News, 44(3):27–39, 2016

    2016

  24. [32]

    PUMA: A programmable ultra-efficient memristor-based accelerator for machine learning inference

    Aayush Ankit, Izzat El Hajj, Sai Rahul Chalamalasetti, Ge- offrey Ndu, Martin Foltin, R Stanley Williams, Paolo Fara- boschi, Wen-mei W Hwu, John Paul Strachan, Kaushik Roy, et al. PUMA: A programmable ultra-efficient memristor-based accelerator for machine learning inference....

    2019

  25. [33]

    A computing-in-memory macro based on three-dimensional resistive random-access memory

    Qiang Huo, Yiming Yang, Yiming Wang, Dengyun Lei, Xi- angqu Fu, Qirui Ren, Xiaoxin Xu, Qing Luo, Guozhong Xing, Chengying Chen, et al. A computing-in-memory macro based on three-dimensional resistive random-access memory. Nature Electronics, 5(7):469–477, 2022

    2022

  26. [34]

    Fast and robust analog in-memory deep neu- ral network training

    Malte J Rasch, Fabio Carta, Omobayode Fagbohungbe, and Tayfun Gokmen. Fast and robust analog in-memory deep neu- ral network training. Nature Communications, 15(1):7133, 2024

    2024

  27. [35]

    The inherent adversarial robustness of analog in-memory computing

    Corey Lammie, Julian B¨ uchel, Athanasios Vasilopoulos, Manuel Le Gallo, and Abu Sebastian. The inherent adversarial robustness of analog in-memory computing. Nature Commu- nications, 16(1):1756, 2025

    2025

  28. [36]

    In-memory computing with resistive memory circuits: Status and outlook

    Giacomo Pedretti and Daniele Ielmini. In-memory computing with resistive memory circuits: Status and outlook. Electron- ics, 10(9):1063, 2021

    2021

  29. [37]

    A crossbar array of magnetoresistive memory devices for in-memory computing

    Seungchul Jung, Hyungwoo Lee, Sungmeen Myung, Hyunsoo Kim, Seung Keun Yoon, Soon-Wan Kwon, Yongmin Ju, Minje Kim, Wooseok Yi, Shinhee Han, et al. A crossbar array of magnetoresistive memory devices for in-memory computing. Nature, 601(7892):211–216, 2022

    2022

  30. [38]

    A 2941-tops/w charge-domain 10t sram compute-in-memory for ternary neural network

    Sungsoo Cheon, Kyeongho Lee, and Jongsun Park. A 2941-tops/w charge-domain 10t sram compute-in-memory for ternary neural network. IEEE Transactions on Circuits and Systems I: Regular Papers, 70(5):2085–2097, 2023

    2023

  31. [39]

    Dou- ble mac on a cell: A 22-nm 8t-sram based analog in-memory 13 accelerator for binary/ternary neural networks featuring split wordline

    Hiroto Tagata, Takashi Sato, and Hiromitsu Awano. Dou- ble mac on a cell: A 22-nm 8t-sram based analog in-memory 13 accelerator for binary/ternary neural networks featuring split wordline. IEEE Open Journal of Circuits and Systems, 2024

    2024

  32. [40]

    Power-efficient combinatorial opti- mization using intrinsic noise in memristor hopfield neural networks

    Fuxi Cai, Suhas Kumar, Thomas Van Vaerenbergh, Xia Sheng, Rui Liu, Can Li, Zhan Liu, Martin Foltin, Shimeng Yu, Qiangfei Xia, et al. Power-efficient combinatorial opti- mization using intrinsic noise in memristor hopfield neural networks. Nature Electronics, 3(7):409–418, 2020

    2020

  33. [41]

    A 64-core mixed-signal in- memory compute chip based on phase-change memory for deep neural network inference

    Manuel Le Gallo, Riduan Khaddam-Aljameh, Milos Stanisavl- jevic, Athanasios Vasilopoulos, Benedikt Kersting, Martino Dazzi, Geethan Karunaratne, Matthias Br¨ andli, Abhairaj Singh, Silvia M Mueller, et al. A 64-core mixed-signal in- memory compute chip based on phase-change me...

    2023

  34. [42]

    Multpim: Fast stateful multiplication for processing-in- memory

    Orian Leitersdorf, Ronny Ronen, and Shahar Kvatinsky. Multpim: Fast stateful multiplication for processing-in- memory. IEEE Transactions on Circuits and Systems II: Ex- press Briefs, 69(3):1647–1651, 2021

    2021

  35. [43]

    Mat- pim: Accelerating matrix operations with memristive stateful logic

    Orian Leitersdorf, Ronny Ronen, and Shahar Kvatinsky. Mat- pim: Accelerating matrix operations with memristive stateful logic. In 2022 IEEE International Symposium on Circuits and Systems (ISCAS), pages 215–219. IEEE, 2022

    2022

  36. [44]

    Magic—memristor-aided logic

    Shahar Kvatinsky, Dmitry Belousov, Slavik Liman, Guy Sa- tat, Nimrod Wald, Eby G Friedman, Avinoam Kolodny, and Uri C Weiser. Magic—memristor-aided logic. IEEE Trans- actions on Circuits and Systems II: Express Briefs, 61(11): 895–899, 2014

    2014

  37. [45]

    Quantized neural networks: Train- ing neural networks with low precision weights and activa- tions

    Itay Hubara, Matthieu Courbariaux, Daniel Soudry, Ran El- Yaniv, and Yoshua Bengio. Quantized neural networks: Train- ing neural networks with low precision weights and activa- tions. journal of machine learning research, 18(187):1–30, 2018

    2018

  38. [46]

    Binarized neural networks

    Itay Hubara, Matthieu Courbariaux, Daniel Soudry, Ran El- Yaniv, and Yoshua Bengio. Binarized neural networks. Ad- vances in neural information processing systems, 29, 2016

    2016

  39. [47]

    Powering ai at the edge: A robust, memristor- based binarized neural network with near-memory computing and miniaturized solar cell

    Fadi Jebali, Atreya Majumdar, Cl´ ement Turck, Kamel- Eddine Harabi, Mathieu-Coumba Faye, Eloi Muhr, Jean- Pierre Walder, Oleksandr Bilousov, Amad´ eo Michaud, Elisa Vianello, et al. Powering ai at the edge: A robust, memristor- based binarized neural network with near-memory ...

    2024

  40. [48]

    Computation over multiple-access channels

    Bobak Nazer and Michael Gastpar. Computation over multiple-access channels. IEEE Transactions on information theory, 53(10):3498–3516, 2007

    2007

  41. [49]

    Robust ana- log function computation via wireless multiple-access chan- nels

    Mario Goldenbaum and Slawomir Stanczak. Robust ana- log function computation via wireless multiple-access chan- nels. IEEE Transactions on Communications, 61(9):3863– 3877, 2013

    2013

  42. [50]

    Nomographic functions: Efficient computation in clustered gaussian sensor networks

    Mario Goldenbaum, Holger Boche, and S lawomir Sta´ nczak. Nomographic functions: Efficient computation in clustered gaussian sensor networks. IEEE Transactions on Wireless Communications, 14(4):2093–2105, 2014

    2014

  43. [51]

    A survey on over-the-air com- putation

    Alphan S ¸ahin and Rui Yang. A survey on over-the-air com- putation. IEEE Communications Surveys & Tutorials, 25(3): 1877–1908, 2023

    1908

  44. [52]

    AirNN: Over-the- air computation for neural networks via reconfigurable intel- ligent surfaces

    Sara Garcia Sanchez, Guillem Reus-Muns, Carlos Bocanegra, Yanyu Li, Ufuk Muncuk, Yousof Naderi, Yanzhi Wang, Stratis Ioannidis, and Kaushik Roy Chowdhury. AirNN: Over-the- air computation for neural networks via reconfigurable intel- ligent surfaces. IEEE/ACM Transactions on N...

    2022

  45. [53]

    AirFC: Designing fully connected layers for neural networks with wireless sig- nals

    Guillem Reus-Muns, Kubra Alemdar, Sara Garcia Sanchez, Debashri Roy, and Kaushik R Chowdhury. AirFC: Designing fully connected layers for neural networks with wireless sig- nals. In Proc. ACM International Symposium on Theory, Al- gorithmic Foundations, and Protocol Design for...

    2023

  46. [54]

    Wireless distributed matrix-vector multiplica- tion using over-the-air computation and analog coding

    Jinho Choi. Wireless distributed matrix-vector multiplica- tion using over-the-air computation and analog coding. IEEE Transactions on Wireless Communications, 23(8):9826–9838, 2024

    2024

  47. [55]

    Uni- versal approximation with xl mimo systems: Ota classi- fication via trainable analog combining

    Kyriakos Stylianopoulos and George C Alexandropoulos. Uni- versal approximation with xl mimo systems: Ota classi- fication via trainable analog combining. arXiv preprint arXiv:2504.12758, 2025

    2025 arXiv

  48. [56]

    Computing functions over-the-air using digital mod- ulations

    Saeed Razavikia, Jos´ e Mairton Barros da Silva, and Carlo Fis- chione. Computing functions over-the-air using digital mod- ulations. In ICC 2023-IEEE International Conference on Communications, pages 5780–5786. IEEE, 2023

    2023

  49. [57]

    A reconfigurable linear rf analog processor for realizing mi- crowave artificial neural network

    Minning Zhu, Tzu-Wei Kuo, and Chung-Tse Michael Wu. A reconfigurable linear rf analog processor for realizing mi- crowave artificial neural network. IEEE Transactions on Mi- crowave Theory and Techniques, 72(2):1290–1301, 2023

    2023

  50. [58]

    An integrated microwave neu- ral network for broadband computation and communication

    Bala Govind, Maxwell G Anderson, Fan O Wu, Peter L McMahon, and Alyssa Apsel. An integrated microwave neu- ral network for broadband computation and communication. Nature Electronics, pages 1–13, 2025

    2025

  51. [59]

    Mi- crowave signal processing using an analog quantum reservoir computer

    Alen Senanian, Sridhar Prabhu, Vladimir Kremenetski, Saswata Roy, Yingkang Cao, Jeremy Kline, Tatsuhiro On- odera, Logan G Wright, Xiaodi Wu, Valla Fatemi, et al. Mi- crowave signal processing using an analog quantum reservoir computer. Nature Communications, 15(1):7490, 2024

    2024

  52. [60]

    A precise four-quadrant multiplier with sub- nanosecond response

    Barrie Gilbert. A precise four-quadrant multiplier with sub- nanosecond response. IEEE journal of solid-state circuits, 3 (4):365–373, 1968

    1968

  53. [61]

    Low voltage per- formance of a microwave CMOS Gilbert cell mixer

    PJ Sulivan, BA Xavier, and Walter H Ku. Low voltage per- formance of a microwave CMOS Gilbert cell mixer. IEEE Journal of Solid-State Circuits, 32(7):1151–1155, 1997

    1997

  54. [62]

    Coaxial frequency mixer, 300–4300 MHz

    Mini-Circuits. Coaxial frequency mixer, 300–4300 MHz. https://www.minicircuits.com/pdfs/ZEM-4300+.pdf

  55. [63]

    Reconsidering os memory optimizations in the presence of disaggregated memory

    Shai Bergman, Priyank Faldu, Boris Grot, Llu ´ ıs Vilanova, and Mark Silberstein. Reconsidering os memory optimizations in the presence of disaggregated memory. In Proceedings of the 2022 ACM SIGPLAN International Symposium on Memory Management, pages 1–14, 2022

    2022

  56. [64]

    Thermal agitation of electric charge in con- ductors

    Harry Nyquist. Thermal agitation of electric charge in con- ductors. Physical review, 32(1):110, 1928

    1928

  57. [65]

    Irreversibility and generalized noise

    Herbert B Callen and Theodore A Welton. Irreversibility and generalized noise. Physical Review, 83(1):34, 1951

    1951

  58. [66]

    Disaggregated deep learning via in-physics computing at radio frequency

    Zhihui Gao, Sri Krishna Vadlamani, Kfir Sulimany, Dirk Englund, and Tingjun Chen. Disaggregated deep learning via in-physics computing at radio frequency. arXiv preprint arXiv:2504.17752, 2025

    2025 arXiv

  59. [67]

    A 0.75-million-point fourier-transform chip for frequency-sparse signals

    Omid Abari, Ezz Hamed, Haitham Hassanieh, Abhinav Agar- wal, Dina Katabi, Anantha P Chandrakasan, and Vladimir Stojanovic. A 0.75-million-point fourier-transform chip for frequency-sparse signals. In Proc. IEEE International Solid- State Circuits Conference (ISSCC), 2014

    2014

  60. [68]

    An IEEE 802.11a/g/p OFDM receiver for GNU Radio

    Bastian Bloessl, Michele Segata, Christoph Sommer, and Falko Dressler. An IEEE 802.11a/g/p OFDM receiver for GNU Radio. In Proc. 2nd Workshop on Software Radio Im- plementation Forum (SRIF), 2013

    2013

  61. [69]

    Ex- perimentally realized in situ backpropagation for deep learn- ing in photonic neural networks

    Sunil Pai, Zhanghao Sun, Tyler W Hughes, Taewon Park, Ben Bartlett, Ian AD Williamson, Momchil Minkov, Maziyar Mi- lanizadeh, Nathnael Abebe, Francesco Morichetti, et al. Ex- perimentally realized in situ backpropagation for deep learn- ing in photonic neural networks. Science...

    2023

  62. [70]

    Large-scale optical neural net- works based on photoelectric multiplication

    Ryan Hamerly, Liane Bernstein, Alexander Sludds, Marin Soljaˇ ci´ c, and Dirk Englund. Large-scale optical neural net- works based on photoelectric multiplication. Physical Review X, 9(2):021032, 2019

    2019

  63. [71]

    RF-photonic deep learning processor with shannon- limited data movement

    Ronald Davis III, Zaijun Chen, Ryan Hamerly, and Dirk En- glund. RF-photonic deep learning processor with shannon- limited data movement. arXiv preprint arXiv:2207.06883v2, 2024

    2024 arXiv

  64. [72]

    Layer normalization

    Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. Layer normalization. arXiv preprint arXiv:1607.06450, 2016

    2016 arXiv

  65. [73]

    Transformers without normalization

    Jiachen Zhu, Xinlei Chen, Kaiming He, Yann LeCun, and Zhuang Liu. Transformers without normalization. In Proceed- ings of the Computer Vision and Pattern Recognition Con- ference, pages 14901–14911, 2025. 14

    2025

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.