Pith. sign in

REVIEW 4 major objections 5 minor 72 references

CN101 - A Digital Thermodynamic Computer for Generative AI

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A digital CMOS chip computes generative AI by averaging random bitstreams.

desk verdict A real digital CMOS prototype with honest measurements, but the central stationarity claim is asserted rather than proved, and the baseline comparisons are softer than the headline numbers suggest. read the letter →

arxiv 2608.00754 v1 pith:GJPFDBJ5 submitted 2026-08-01 cs.ET cs.AI

Lars Holdijk , Denis Melanson , Zier Mensch , Brandon Birchall , Vincent Cheung , Nicholas Lehrter , Maxwell Aifer , Samuel Duffield
show 9 more authors
Jan Ole Ernst Rajath Salegame Antonio J. Martinez Gavin Crooks Miranda Cheng Zach Belateche Marc Bright Patrick J. Coles Faris Sbahi
This is my paper · ORCID
classification cs.ETcs.AI
keywords thermodynamiccomputingequilibrationformalismstochasticgenerativeAIflowmatchingsequentialparallelismanytimeprecisionfree-energyestimation
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

The paper tries to establish that the equilibration style of thermodynamic computing—evaluating a function as the stationary expectation of an ergodic stochastic process—is not tied to analogue Langevin hardware. It gives a substrate-independent definition of such a computation as a tuple $(S,L,\pi,\varphi)$ with $f(x)=E_\pi[\varphi(S)]$, where $L$ is any ergodic dynamical generator. The authors then instantiate this definition on CN101, a digital CMOS test chip whose accumulator-based stochastic-computing dynamics form a discrete Markov chain. On that chip they demonstrate digital thermodynamic linear algebra, a conditional VAE on handwritten digits, flow-matching image generation, and free-energy estimation on alanine dipeptide, matching floating-point references of the same quantized models. If the central claim holds, thermodynamic computing's hardware benefits—anytime precision, parallel sample aggregation, and sequential parallelism—are available on ordinary digital silicon rather than only on analogue substrates.

What carries the argument

The load-bearing object is the equilibration tuple $(S,L,\pi,\varphi)$ together with the time-averaged readout equation. $L$ is the generator of an ergodic Markov chain; the paper's key move is that $L$ may be a discrete accumulator-based stochastic-computing datapath rather than a continuous Langevin operator. On CN101 the machinery consists of 8-bit deterministic weights gated by AND gates against stochastic input bitstreams, OR-based column accumulation, split-unipolar encoding for signed arithmetic, per-neuron 65-bit LFSR pseudo-random generators, a deterministic multi-bit path inside each tile, and the stochastic streaming network-on-chip that carries bitstreams between tiles on independent clocks. The analytical engine for sequential parallelism is the continuous-mapping theorem: if each layer map is continuous, a downstream stage consuming a still-converging estimate stays consistent, so the pipeline converges almost surely to the composed function as the joint relaxation proceeds.

What would settle it

Program a single tile with known weights and bias, choose an activation that is neither linear nor ReLU (for example the finite-state-machine sigmoid path), run to a long averaging window, and compare the chip's time-averaged readout against the exact $\sigma(Wx+b)$ computed in floating point; a systematic offset that persists beyond the 8-bit weight-quantisation floor would falsify the claim that the chip's stationary expectation equals the intended function.

Watch

Extended reading notes

Core claim

The paper's central claim is that a thermodynamic computation is fully specified by the dynamical generator $L$ of an ergodic process, not by any specific physics. On a state space $S$, with generator $L$ parametrized by the input $x$, unique stationary distribution $\pi$ satisfying $L\pi=0$, and readout $\varphi$, the computed function is $f(x)=E_\pi[\varphi(S)]$, returned as the time-averaged readout $\hat y_T=\frac{1}{T}\sum_t \varphi(S_t)$, which converges almost surely as $T\to\infty$. The Ornstein–Uhlenbeck dynamics of prior analogue work are one special case; a discrete-state Markov chain realized in digital logic is another. The paper reports that CN101 embodies the second case: four tiles of $64\times 64$ stochastic matrix multipliers, reconfigurable neuron banks, and a stochastic streaming network-on-chip form a joint Markov chain whose stationary expectation is the composed network function. The paper's Eq. (5) claims, via the continuous-mapping theorem, that an $L$-stage pipeline of continuous layer maps converges almost surely to the composed function $(f_L\circ\cdots\circ f_1)(x)$ even when downstream stages consume still-converging upstream estimates, which is the basis for running dependent stages concurrently.

Load-bearing premise

The argument assumes that for every implemented layer the stochastic-computing accumulator dynamics settle to a long-run average exactly equal to the layer's mathematical output; if a nonlinearity or saturation effect makes that average come out differently, the chip would converge to the wrong function.

Editorial extensions

If this is right

  • Precision becomes a runtime knob: a single run of the chip returns a usable coarse estimate at short windows and refines it monotonically as the averaging window grows, so workloads can stop once the downstream task's accuracy is reached.
  • Independent trajectories are statistically independent, so averaging $K$ parallel replicas reduces the estimator variance by a factor $K$ with no inter-replica coordination, although the burn-in bias is shared across replicas and cannot be averaged away.
  • Dependent stages of a deep generative model can relax concurrently: in the CIFAR-10 flow-matching experiment the unrolled 100-layer integration reached the floating-point reference at about $10^7$ cycles per pixel, versus roughly $6.2\times 10^8$ cycles when layers were run sequentially, a factor near 62 at that accuracy target.
  • The formulation is substrate-independent, so future instantiations may use any ergodic dynamics—analogue, digital, or hybrid—whose stationary distribution is programmable, removing the CMOS toolchain restriction that analogue thermodynamic hardware faces.
  • On a molecular benchmark, the chip's free-energy estimates agree with the CPU estimate of the same flows to within $0.8$ kJ/mol and track an umbrella-sampling reference for five of the six alanine-dipeptide basins.

Reading between the lines

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

  • One extension the paper does not itself run is a stress test of the stationarity assumption on a generic finite-state-machine nonlinearity such as sigmoid or tanh; the chip contains an FSM activation path, and a direct long-window readout comparison would show whether the accumulator dynamics' stationary mean equals the intended function for non-ReLU layers.
  • If the same joint-relaxation argument transfers to autoregressive language models, token generation could be organized as a single relaxation over token positions, each consuming still-converging estimates of earlier positions, so wall-clock could scale sublinearly in sequence length; the paper does not attempt this extension.
  • Building on the paper's stated preference for learned non-equilibrium protocols, one testable extension is to drive the accumulator chain through a finite-time schedule and see whether the burn-in floor that limits parallel sample aggregation is reduced.
  • The reported bias floor is attributed to 8-bit weight-register precision, which suggests a testable hardware lever: increasing register width or adding a readout correction should lower the floor on all workloads, since the paper identifies quantisation rather than the formalism as the cause.
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

4 major / 5 minor

Summary. The manuscript proposes a substrate-independent formalization of equilibrium thermodynamic computing, defined by a tuple (S,L,π,φ) with f(x)=E_π[φ(S)], and claims three hardware-level scaling properties: anytime precision, parallel sample aggregation, and sequential parallelism. It then presents CN101, a fabricated digital CMOS chip that implements the formalism using stochastic-computing accumulator dynamics and AND/OR arithmetic, and reports experimental results on digital thermodynamic linear algebra, a conditional VAE on MNIST, a convolutional flow-matching model on CIFAR-10, and free-energy estimation on alanine dipeptide across multiple chips. The central claims are that the chip's dynamics are an ergodic Markov chain whose stationary expectation equals the intended network function, and that concurrent execution of dependent layers converges to the composed function via Eq. (5).

Significance. If the central claims hold, this is a significant contribution: it would decouple the equilibration-style thermodynamic computing formulation from analogue Langevin hardware and demonstrate a digital CMOS instantiation with concrete scaling advantages for generative AI workloads. The formalism is clearly stated and is not circular, since f(x) is defined from a designer-chosen generator and the chip outputs are compared against reference computations. The paper also deserves credit for fabricating a prototype chip, running standard generative models without custom layers, reporting explicit bias floors and burn-in limitations, and providing extensive experimental details in appendices. The significance is conditional, however, on resolving the load-bearing gap identified below: the stationarity of the stochastic-computing datapath is asserted rather than proved or validated on a known-answer test.

major comments (4)
  1. [Section 3.1.1 / 3.2] The paper asserts, in Section 3.2, that the weights, biases, and activations loaded onto the chip fix a stationary distribution π satisfying Lπ=0 and that the time-averaged readout converges to E_π[φ(S)] = f(x). The transition kernel of the actual stochastic-computing datapath is never written down, and the assertion is not supported for the AND/OR arithmetic. Section 3.1.1 states that OR accumulation 'saturates to mean 1−∏(1−p_i), which has to be corrected at readout,' but no unbiased correction is specified. Applying a nonlinear inverse to a time average is generally biased, so even as T→∞ the readout may not equal the intended layer function σ(Wx+b). Because every generative workload relies on this datapath, this is a load-bearing gap rather than a presentation issue.
  2. [Section 4.1] The only known-answer validation on the chip is the thermodynamic linear algebra experiment, but Section 4.1 explicitly states that it exercises 'a different datapath from the stochastic-computing equilibration the generative models use,' namely linear-integration OU dynamics. This experiment therefore cannot confirm that the AND/OR SMM path, the 8-bit deterministic weight gating, the OR saturation readout, or the FSM nonlinearities have stationary expectations equal to the intended functions. A known-answer test on the stochastic-computing path itself, such as exact verification of a random affine-plus-nonlinear layer against a floating-point computation for the same inputs, is needed before the central claim of Section 3.2 can be accepted.
  3. [Section 2.3.3, Eq. (5)] The sequential-parallelism guarantee is asserted through the continuous-mapping theorem without proving that the coupled joint Markov chain is ergodic or that its stationary expectation is the composed function (f_L∘⋯∘f_1)(x). The argument requires each stage's time-averaged readout to converge to f_ℓ of the current upstream estimate, which is exactly the property in question; Appendix A defines this as an assumption ('assume each stage is consistent'), effectively restating the claim. The authors need to provide conditions under which the joint chain's stationary distribution yields the composed expectation, or a direct proof that the stage-wise consistency plus continuity of the f_ℓ is sufficient in this coupled, non-independent setting.
  4. [Appendix C / Fig. 6(c)] The anytime-precision demonstration selects a different operating point for each T by searching over the chip's input-scale and weight-precision knobs (Appendix C). These knobs are free parameters that can absorb systematic datapath bias, so the reported RMSE-versus-T curve does not isolate the anytime property of the hardware. Additionally, the generative reference is the same 8-bit-quantized model run in floating point (Appendix F), which tests consistency with a chosen quantization rather than correctness against independent ground truth; this is a meaningful but weaker claim than 'the chip computes the intended function.'
minor comments (5)
  1. [Abstract / Section 2.2] The abstract uses L* for the dynamical generator while Section 2.2 defines L; the notation should be unified.
  2. [Appendix E] The symbol T is used both for temperature (T=300 K) and for readout window length; this overloading is confusing and should be resolved, for example by using β for inverse temperature.
  3. [Fig. 3(b) caption] The caption says the OR gate 'accumulates a column of products,' but an OR gate is not an accumulator in the usual stochastic-computing sense; the caption should clarify that the time average of the OR output is what accumulates the saturation-corrected sum.
  4. [References] Reference [23] appears incomplete as printed; the author list and full bibliographic details should be supplied.
  5. [Fig. 8] The caption of Fig. 8 does not define the color scale or the units of per-tile error; these should be stated explicitly.

Circularity Check

1 steps flagged · score 3.0 of 10

Anytime-precision curve is partly a fitted lower envelope; the formalism and fixed-configuration checks are otherwise self-contained.

  1. fitted input called prediction [Section 4.2.2 (Anytime-precision) and Appendix C]
    "Fig. 6c plots the RMSE as a function of T, with per-T operating points selected by a search over the chip's scale and precision knobs (details in Appendix C). ... For each cycle budget T the configuration minimising the scale-invariant RMSE against the reference is chosen, so that the reported curve traces the best operating point available at each T rather than a single fixed setting."

    The anytime-precision demonstration is presented as evidence of convergence with averaging window, but the RMSE at each T is not produced by a fixed chip configuration: it is the minimum over input-scale and weight-register-precision settings chosen using the CPU reference. The improvement in the curve is therefore partly a property of the per-T search, not purely of longer averaging, so the 'predicted' anytime-precision scaling is partly constructed by fitting the knobs to the target. This does not invalidate the formal anytime-precision property, which follows from the ergodic theorem, nor the fixed-configuration chip-versus-reference checks, but the specific RMSE-versus-T curve is, to that extent, self-fulfilling.

full rationale

The central formalism of Section 2.2 is definitional rather than circular: f(x) = E_pi[phi(S)] is a definition of what a thermodynamic computation is, and Eq. (5) follows from the continuous-mapping theorem applied to stage-wise consistent estimators, not from an assumption equivalent to the conclusion. The chip's generative results are compared against a floating-point reference of the same quantized model, so the substrate's correctness is externally checked in a fixed-configuration sense. Self-citations (TLA, SPU, free-energy protocol) provide background and protocol choices; they are not used to import a uniqueness theorem or to forbid alternative generators. No ansatz is smuggled in via citation: the AND/OR stochastic-computing arithmetic is standard and its stationary means are elementary. The only identified circular element is the anytime-precision plot, whose per-T points are selected by minimizing RMSE against the reference, making the reported curve a best-case envelope rather than a prediction at a fixed setting. Because this affects a supporting demonstration rather than the load-bearing derivation of the paper's central claim, the appropriate score is a moderate-low 3.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

No new physical entities are postulated. CN101 is a physical prototype and the Carnot Architecture is a design framework realized by the chip, not an invented entity used to explain results. The ledger's free parameters are the experimental tuning and scheduling choices; the axioms are the assumptions under which the formalism and the chip's stochastic datapath are claimed to compute correctly.

free parameters (4)
  • VAE per-T input scale and weight-precision knobs = not reported (searched per T)
    Appendix C: for each cycle budget T the configuration minimizing scale-invariant RMSE against the CPU reference is chosen, so the anytime-precision curve is an optimized envelope, not a fixed setting.
  • Readout window schedule for CIFAR and alanine = rising to 5e5 and 1e7 cycles per tile respectively
    Chosen by hand; the schedule sets the anytime-precision trade-off and burn-in discard; no sensitivity analysis is given.
  • Weight quantization to 8-bit = 8-bit
    Chosen for deployment; sets the bias floor reported (roughly 1e-2 relative error).
  • Multi-chip sync interval = not reported for experiments
    The host refreshes downstream inputs window by window; the chosen interval determines how close the system is to the fully concurrent limit, but the value used in Section 5.2 is not stated.
assumptions (4)
  • domain assumption Ergodic dynamics with unique stationary distribution exist for the chosen generators.
    Section 2.2 defines the formalism on this assumption, citing Meyn and Tweedie; whether CN101's finite-state machine satisfies it is asserted rather than proven.
  • domain assumption LFSR pseudo-random streams behave as independent Bernoulli bits.
    Section 3.1.1 and 3.1.3 assume bitstream means are recovered as time-averages; LFSR bits are deterministic and correlated, which the paper does not analyze beyond period greater than 1e19.
  • standard math Stage maps f_l are continuous, enabling the continuous-mapping theorem in sequential parallelism.
    Section 2.3.3 and Appendix A rely on continuity of each stage and the extended continuous-mapping theorem; the theorem itself is standard, but the applicability to coupled running estimates is not proven.
  • ad hoc to paper The stochastic datapath's stationary expectation equals the intended affine-plus-nonlinear function.
    Section 3.2 asserts the chip's state has stationary distribution π with Lπ=0 encoding the network, but no derivation is given for AND/OR arithmetic with deterministic 8-bit weights and readout correction.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CN101 - A Digital Thermodynamic Computer for Generative AI." pith.science (2026). https://pith.science/paper/GJPFDBJ5

@misc{pith2026260800754,
  author       = {Pith},
  title        = {Pith review of: CN101 - A Digital Thermodynamic Computer for Generative AI},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GJPFDBJ5}},
  note         = {Machine review of arXiv:2608.00754}
}
read the original abstract

Thermodynamic computing is an emerging hardware paradigm, in which stochastic physical dynamics serve as the direct computational primitive. The recent explosion of generative AI has only sharpened the search for alternative approaches to compute, and, as we show in this work, thermodynamic computing turns out to be well suited to this space. An important class of methods realises a function as the stationary expectation of an ergodic stochastic process: the answer is encoded in the time-averaged statistics of an equilibrating trajectory. To date, this equilibration-style class has been formulated exclusively through Langevin dynamics, restricting its implementations to analogue substrates and the engineering challenges those bring. In this work, we propose a substrate-independent formalisation of the equilibration-style formulation, in which the only object of design is the dynamical generator L* of an arbitrary ergodic process. The formalisation makes three hardware-level properties of the formulation explicit: the precision of a result is a knob set by how long the dynamics are run, sample averages decompose across independent trajectories, and dependent stages of a computation operate concurrently rather than serially, a property we call sequential parallelism. We instantiate the formalisation by fabricating a prototype digital thermodynamic computing chip, named CN101, that implements the formulation through discrete accumulator dynamics on standard CMOS using stochastic computing principles. We characterise CN101's success across conventional generative AI workloads in the form of VAEs and flow matching, applied to both image generation and scientific problems. Together, the formalisation and its digital instantiation show that the equilibration-style formulation is substrate-independent, and that its computational properties can be exploited on standard digital hardware.

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

72 extracted references · 36 canonical work pages

  1. [1]

    Tom Conte, Erik DeBenedictis, Natesh Ganesh, Todd Hyl- ton, John Paul Strachan, R. Stanley Williams, Alexander Alemi, Lee Altenberg, Gavin Crooks, James Crutchfield, Lidia del Rio, Josh Deutsch, Michael DeWeese, Khari Dou- glas, Massimiliano Esposito, Michael Frank, Robert Fry, Peter Harsha, Mark Hill, Christopher Kello, Jeff Krichmar, Suhas Kumar, Shih-C...

    arXiv 2019

  2. [2]

    Coles, Collin Szczepanski, Denis Melanson, Kae- lan Donatella, Antonio J

    Patrick J. Coles, Collin Szczepanski, Denis Melanson, Kae- lan Donatella, Antonio J. Martinez, and Faris Sbahi. Ther- modynamic AI and the fluctuation frontier. In2023 IEEE International Conference on Rebooting Computing (ICRC), pages 1–10. IEEE, 2023. doi: 10.1109/ICRC60800.2023. 10386858

    arXiv

  3. [3]

    Thermodynamic computing: An intellectual and technological frontier.Proceedings, 47(1):23, 2020

    Todd Hylton. Thermodynamic computing: An intellectual and technological frontier.Proceedings, 47(1):23, 2020. doi: 10.3390/proceedings2020047023. MORCOM 2019

    work page doi:10.3390/proceedings2020047023 2020

  4. [4]

    Camsari, Patrick J

    Maxwell Aifer, Zach Belateche, Suraj Bramhavar, Kerem Y . Camsari, Patrick J. Coles, Gavin Crooks, Douglas J. Durian, Andrea J. Liu, Anastasia Marchenkova, Antonio J. Mar- tinez, Peter L. McMahon, Faris Sbahi, Benjamin Weiner, and Logan G. Wright. Solving the compute crisis with physics-based ASICs, 2025. URL https://arxiv.org/ abs/2507.10463

    arXiv 2025

  5. [5]

    Maxwell Aifer, Kaelan Donatella, Max Hunter Gordon, Samuel Duffield, Thomas Ahle, Daniel Simpson, Gavin Crooks, and Patrick J. Coles. Thermodynamic linear alge- bra.npj Unconventional Computing, 1(1):13, 2024. doi: 10.1038/s44335-024-00014-0

    doi:10.1038/s44335-024-00014-0 2024

  6. [6]

    Crooks, Antonio J

    Denis Melanson, Mohammad Abu Khater, Maxwell Aifer, Kaelan Donatella, Max Hunter Gordon, Thomas Ahle, Gavin E. Crooks, Antonio J. Martinez, Faris Sbahi, and Patrick J. Coles. Thermodynamic computing system for AI applications.Nature Communications, 16(1):3757, 2025. doi: 10.1038/s41467-025-59011-x

    doi:10.1038/s41467-025-59011-x 2025

  7. [7]

    Martinez, and Patrick J

    Maxwell Aifer, Samuel Duffield, Kaelan Donatella, Denis Melanson, Phoebe Klett, Zach Belateche, Gavin Crooks, Antonio J. Martinez, and Patrick J. Coles. Thermodynamic Bayesian inference. In2024 IEEE International Confer- ence on Rebooting Computing (ICRC), pages 1–20. IEEE,

    work page

  8. [8]

    Kaelan Donatella, Samuel Duffield, Maxwell Aifer, Denis Melanson, Gavin Crooks, and Patrick J. Coles. Thermody- namic natural gradient descent.npj Unconventional Com- puting, 3(1):5, 2026. doi: 10.1038/s44335-025-00049-x

    work page doi:10.1038/s44335-025-00049-x 2026

Show all 72 references
  1. [9]

    Nonlinear ther- modynamic computing out of equilibrium.Nature Communications, 17(1):1189, 2026

    Stephen Whitelam and Corneel Casert. Nonlinear ther- modynamic computing out of equilibrium.Nature Communications, 17(1):1189, 2026. doi: 10.1038/ s41467-025-67958-0

    2026

  2. [10]

    Generative thermodynamic computing

    Stephen Whitelam. Generative thermodynamic computing. Physical Review Letters, 136(3):037101, 2026. doi: 10. 1103/kwyy-1xln

    2026

  3. [11]

    Camsari, Rafatul Faria, Brian M

    Kerem Y . Camsari, Rafatul Faria, Brian M. Sutton, and Supriyo Datta. Stochastic p-bits for invertible logic.Physi- cal Review X, 7(3):031014, 2017. doi: 10.1103/PhysRevX. 7.031014

    2017 doi

  4. [12]

    Probabilistic computing with p-bits.Applied Physics Letters, 119(15):150503, 2021

    Jan Kaiser and Supriyo Datta. Probabilistic computing with p-bits.Applied Physics Letters, 119(15):150503, 2021. doi: 10.1063/5.0067927

    2021 doi

  5. [13]

    Martinis, Giovanni Finocchio, and Kerem Y

    Navid Anjum Aadit, Andrea Grimaldi, Mario Carpentieri, Luke Theogarajan, John M. Martinis, Giovanni Finocchio, and Kerem Y . Camsari. Massively parallel probabilistic computing with sparse Ising machines.Nature Electronics, 5(7):460–468, 2022. doi: 10.1038/s41928-022-00774-2

    2022 doi

  6. [14]

    An efficient probabilistic hardware architecture for diffusion-like models, 2025

    Andraž Jelinˇciˇc, Owen Lockwood, Akhil Garlapati, Peter Schillinger, Isaac Chuang, Guillaume Verdon, and Trevor McCourt. An efficient probabilistic hardware architecture for diffusion-like models, 2025. URL https://arxiv. org/abs/2510.23972

    2025

  7. [15]

    Brian R. Gaines. Stochastic computing systems. In Julius T. Tou, editor,Advances in Information Systems Science, Vol. 2, pages 37–172. Plenum Press, 1969

    1969

  8. [16]

    Armin Alaghi and John P. Hayes. Survey of stochastic computing.ACM Transactions on Embedded Computing Systems, 12(2s):92:1–92:19, 2013. doi: 10.1145/2465787. 2465794

    2013 doi

  9. [17]

    G. E. Uhlenbeck and L. S. Ornstein. On the theory of the Brownian motion.Physical Review, 36(5):823–841, 1930. doi: 10.1103/PhysRev.36.823

    1930 doi

  10. [18]

    Meyn and Richard L

    Sean P. Meyn and Richard L. Tweedie.Markov Chains and Stochastic Stability. Cambridge University Press, 2nd edition, 2009. ISBN 978-0-521-73182-9. CN101 - A Digital Thermodynamic Computer for Generative AI 16

    2009

  11. [19]

    Jones, and Xiao- Li Meng, editors.Handbook of Markov Chain Monte Carlo

    Steve Brooks, Andrew Gelman, Galin L. Jones, and Xiao- Li Meng, editors.Handbook of Markov Chain Monte Carlo. Handbooks of Modern Statistical Methods. Chapman & Hall / CRC, 2011. ISBN 978-1-4200-7941-8

    2011

  12. [20]

    Using anytime algorithms in intelligent systems.AI Magazine, 17(3):73–83, 1996

    Shlomo Zilberstein. Using anytime algorithms in intelligent systems.AI Magazine, 17(3):73–83, 1996. doi: 10.1609/ aimag.v17i3.1232

    1996

  13. [21]

    A. W. van der Vaart.Asymptotic Statistics. Cam- bridge Series in Statistical and Probabilistic Mathemat- ics. Cambridge University Press, 1998. doi: 10.1017/ CBO9780511802256

    1998

  14. [22]

    parareal

    Jacques-Louis Lions, Yvon Maday, and Gabriel Turinici. A “parareal” in time discretization of PDE’s.Comptes Rendus de l’Académie des Sciences - Series I - Mathematics, 332 (7):661–668, 2001. doi: 10.1016/S0764-4442(00)01793-6

    2001 doi

  15. [23]

    Parallel sampling of diffusion models

    Andy Shih, Suneel Belkhale, Stefano Ermon, Dorsa Sadigh, and Nima Anari. Parallel sampling of diffusion models. InAdvances in Neural Information Processing Systems (NeurIPS), 2023

    2023

  16. [24]

    Hardware ODE solvers using stochastic circuits

    Siting Liu and Jie Han. Hardware ODE solvers using stochastic circuits. InDesign Automation Conference (DAC), 2017. doi: 10.1145/3061639.3062258

    2017

  17. [25]

    Gross, and Jie Han

    Siting Liu, Warren J. Gross, and Jie Han. Introduction to dynamic stochastic computing.IEEE Circuits and Systems Magazine, 20(3):19–33, 2020. doi: 10.1109/MCAS.2020. 3005483

    2020 doi

  18. [26]

    ACOUSTIC: Acceler- ating convolutional neural networks through or-unipolar skipped stochastic computing

    Wojciech Romaszkan, Tianmu Li, Tristan Melton, Sud- hakar Pamarti, and Puneet Gupta. ACOUSTIC: Acceler- ating convolutional neural networks through or-unipolar skipped stochastic computing. InDesign, Automation & Test in Europe Conference & Exhibition (DATE), pages 768–773, 20...

    2020

  19. [27]

    SC-DCNN: Highly- scalable deep convolutional neural network using stochas- tic computing

    Ao Ren, Zhe Li, Caiwen Ding, Qinru Qiu, Yanzhi Wang, Ji Li, Xuehai Qian, and Bo Yuan. SC-DCNN: Highly- scalable deep convolutional neural network using stochas- tic computing. InInternational Conference on Archi- tectural Support for Programming Languages and Op- erating Syste...

    2017

  20. [28]

    Dynamic energy- accuracy trade-off using stochastic computing in deep neu- ral networks

    Kyounghoon Kim, Jungki Kim, Joonsang Yu, Jungwoo Seo, Jongeun Lee, and Kiyoung Choi. Dynamic energy- accuracy trade-off using stochastic computing in deep neu- ral networks. InDesign Automation Conference (DAC),

  21. [29]

    Scalable stochastic-computing accelerator for convo- lutional neural networks

    Hyeonuk Sim, Dong Nguyen, Jongeun Lee, and Kiyoung Choi. Scalable stochastic-computing accelerator for convo- lutional neural networks. InAsia and South Pacific Design Automation Conference (ASP-DAC), pages 696–701, 2017. doi: 10.1109/ASPDAC.2017.7858405

    2017

  22. [30]

    Brown and Howard C

    Bradley D. Brown and Howard C. Card. Stochastic neural computation I: Computational elements.IEEE Transactions on Computers, 50(9):891–905, 2001. doi: 10.1109/12.954505

    2001 doi

  23. [31]

    Hardware-driven nonlinear activation for stochastic computing based deep convolutional neural networks

    Ji Li, Zihao Yuan, Zhe Li, Caiwen Ding, Ao Ren, Qinru Qiu, Jeffrey Draper, and Yanzhi Wang. Hardware-driven nonlinear activation for stochastic computing based deep convolutional neural networks. InInternational Joint Con- ference on Neural Networks (IJCNN), pages 1230–1236, 2017

    2017

  24. [32]

    Florian Neugebauer, Ilia Polian, and John P. Hayes. S-box- based random number generation for stochastic computing. Microprocessors and Microsystems, 61:316–326, 2018. doi: 10.1016/j.micpro.2018.06.009

    2018 doi

  25. [33]

    Chapiro.Globally-Asynchronous Locally- Synchronous Systems

    Daniel M. Chapiro.Globally-Asynchronous Locally- Synchronous Systems. PhD thesis, Stanford University,

  26. [34]

    Samuel Duffield, Maxwell Aifer, Denis Melanson, Zach Belateche, and Patrick J. Coles. Lattice random walk dis- cretisations of stochastic differential equations, 2025

    2025

  27. [35]

    Coles, Max Welling, and Miranda C

    Zier Mensch, Lars Holdijk, Samuel Duffield, Maxwell Aifer, Patrick J. Coles, Max Welling, and Miranda C. N. Cheng. Robust stochastic gradient posterior sampling with lattice based discretisation, 2026

    2026

  28. [36]

    Kingma and Max Welling

    Diederik P. Kingma and Max Welling. Auto-encoding variational bayes. InInternational Conference on Learning Representations (ICLR), 2014

    2014

  29. [37]

    Learning structured output representation using deep conditional generative models

    Kihyuk Sohn, Honglak Lee, and Xinchen Yan. Learning structured output representation using deep conditional generative models. InAdvances in Neural Information Processing Systems (NeurIPS), pages 3483–3491, 2015

    2015

  30. [38]

    Optical recognition of handwritten digits

    Ethem Alpaydin and Cenk Kaynak. Optical recognition of handwritten digits. UCI Machine Learning Repository, 1998

    1998

  31. [39]

    Gradient-based learning applied to document recognition.Proceedings of the IEEE, 86(11):2278–2324,

    Yann LeCun, Léon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to document recognition.Proceedings of the IEEE, 86(11):2278–2324,

  32. [40]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. InIEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016

    2016

  33. [41]

    The hardware lottery.Communications of the ACM, 64(12):58–65, 2021

    Sara Hooker. The hardware lottery.Communications of the ACM, 64(12):58–65, 2021. doi: 10.1145/3467017. CN101 - A Digital Thermodynamic Computer for Generative AI 17

    2021 doi

  34. [42]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. InAdvances in Neural Information Processing Systems (NeurIPS), 2020

    2020

  35. [43]

    Weiss, Niru Mah- eswaranathan, and Surya Ganguli

    Jascha Sohl-Dickstein, Eric A. Weiss, Niru Mah- eswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. InInter- national Conference on Machine Learning (ICML), 2015

    2015

  36. [44]

    A complete decomposition of stochastic differential equations, 2026

    Samuel Duffield. A complete decomposition of stochastic differential equations, 2026

    2026

  37. [45]

    Ricky T. Q. Chen, Yulia Rubanova, Jesse Bettencourt, and David Duvenaud. Neural ordinary differential equations. InAdvances in Neural Information Processing Systems (NeurIPS), 2018

    2018

  38. [46]

    A proposal on machine learning via dynamical systems.Communications in Mathematics and Statistics, 5 (1):1–11, 2017

    Weinan E. A proposal on machine learning via dynamical systems.Communications in Mathematics and Statistics, 5 (1):1–11, 2017. doi: 10.1007/s40304-017-0103-z

    2017 doi

  39. [47]

    Kloeden and Eckhard Platen.Numerical Solu- tion of Stochastic Differential Equations, volume 23 of Applications of Mathematics

    Peter E. Kloeden and Eckhard Platen.Numerical Solu- tion of Stochastic Differential Equations, volume 23 of Applications of Mathematics. Springer, 1992. ISBN 3-540- 54062-8

    1992

  40. [48]

    Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole

    Yang Song, Jascha Sohl-Dickstein, Diederik P. Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score- based generative modeling through stochastic differential equations. InInternational Conference on Learning Repre- sentations (ICLR), 2021

    2021

  41. [49]

    Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maxi- milian Nickel, and Matt Le. Flow matching for generative modeling. InInternational Conference on Learning Repre- sentations (ICLR), 2023

    2023

  42. [50]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky. Learning multiple layers of features from tiny images. Technical report, University of Toronto, 2009

    2009

  43. [51]

    Shift: A zero FLOP, zero parameter alternative to spatial convolutions

    Bichen Wu, Alvin Wan, Xiangyu Yue, Peter Jin, Sicheng Zhao, Noah Golmant, Amir Gholaminejad, Joseph Gonza- lez, and Kurt Keutzer. Shift: A zero FLOP, zero parameter alternative to spatial convolutions. InIEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018

    2018

  44. [52]

    U- Net: Convolutional networks for biomedical image seg- mentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U- Net: Convolutional networks for biomedical image seg- mentation. InMedical Image Computing and Computer- Assisted Intervention (MICCAI), pages 234–241, 2015. doi: 10.1007/978-3-319-24574-4_28

    2015 doi

  45. [53]

    Flow straight and fast: Learning to generate and transfer data with rectified flow

    Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow. InInternational Conference on Learn- ing Representations (ICLR), 2023

    2023

  46. [54]

    Boltzmann generators: Sampling equilibrium states of many-body systems with deep learning.Science, 365 (6457):eaaw1147, 2019

    Frank Noé, Simon Olsson, Jonas Köhler, and Hao Wu. Boltzmann generators: Sampling equilibrium states of many-body systems with deep learning.Science, 365 (6457):eaaw1147, 2019. doi: 10.1126/science.aaw1147

    2019 doi

  47. [55]

    Bronstein, and Max Welling

    Lars Holdijk, Nithishwer Mouroug Anand, Michael M. Bronstein, and Max Welling. Learning escorted proto- cols for multistate free-energy estimation. InThe Four- teenth International Conference on Learning Representa- tions, 2026. URL https://openreview.net/forum? id=Da8PJXp0js

    2026

  48. [56]

    Nonequilibrium equality for free energy differences.Physical Review Letters, 78(14):2690– 2693, 1997

    Christopher Jarzynski. Nonequilibrium equality for free energy differences.Physical Review Letters, 78(14):2690– 2693, 1997. doi: 10.1103/PhysRevLett.78.2690

    1997 doi

  49. [57]

    Jarzynski

    C. Jarzynski. Targeted free energy perturbation.Physical Review E, 65(4):046122, 2002. doi: 10.1103/PhysRevE.65. 046122

    2002 doi

  50. [58]

    Ricky T. Q. Chen and Yaron Lipman. Flow matching on general geometries. InInternational Conference on Learning Representations (ICLR), 2024

    2024

  51. [59]

    Charles H. Bennett. Efficient estimation of free energy differences from Monte Carlo data.Journal of Compu- tational Physics, 22(2):245–268, 1976. doi: 10.1016/ 0021-9991(76)90078-4

    1976

  52. [60]

    Will Grathwohl, Ricky T. Q. Chen, Jesse Bettencourt, Ilya Sutskever, and David Duvenaud. FFJORD: Free-form con- tinuous dynamics for scalable reversible generative models. InInternational Conference on Learning Representations (ICLR), 2019

    2019

  53. [61]

    G. M. Torrie and J. P. Valleau. Nonphysical sampling distributions in Monte Carlo free-energy estimation: Um- brella sampling.Journal of Computational Physics, 23(2): 187–199, 1977. doi: 10.1016/0021-9991(77)90121-8

    1977 doi

  54. [62]

    Robert W. Zwanzig. High-temperature equation of state by a perturbation method. I. nonpolar gases.The Journal of Chemical Physics, 22(8):1420–1426, 1954. doi: 10.1063/1. 1740409

    1954 doi

  55. [63]

    Levin, Yuval Peres, and Elizabeth L

    David A. Levin, Yuval Peres, and Elizabeth L. Wilmer. Markov Chains and Mixing Times. American Mathematical Society, 2nd edition, 2017. ISBN 978-1-4704-2962-1

    2017

  56. [64]

    Gavin E. Crooks. Entropy production fluctuation theorem and the nonequilibrium work relation for free energy dif- ferences.Physical Review E, 60(3):2721–2726, 1999. doi: 10.1103/PhysRevE.60.2721

    1999 doi

  57. [65]

    Stochastic thermodynamics, fluctuation the- orems and molecular machines.Reports on Progress in Physics, 75(12):126001, 2012

    Udo Seifert. Stochastic thermodynamics, fluctuation the- orems and molecular machines.Reports on Progress in Physics, 75(12):126001, 2012. doi: 10.1088/0034-4885/ 75/12/126001. CN101 - A Digital Thermodynamic Computer for Generative AI 18

    2012 doi

  58. [66]

    Sivak and Gavin E

    David A. Sivak and Gavin E. Crooks. Thermodynamic metrics and optimal paths.Physical Review Letters, 108 (19):190602, 2012. doi: 10.1103/PhysRevLett.108.190602

    2012 doi

  59. [67]

    Gavin E. Crooks. Measuring thermodynamic length.Phys- ical Review Letters, 99(10):100602, 2007. doi: 10.1103/ PhysRevLett.99.100602

    2007

  60. [68]

    M. F. Hutchinson. A stochastic estimator of the trace of the influence matrix for Laplacian smoothing splines.Commu- nications in Statistics - Simulation and Computation, 18 (3):1059–1076, 1989. doi: 10.1080/03610918908812806. Appendix A. Formal statement of the equilibration ...

    1989 doi

  61. [1984]

    Technical Report STAN-CS-84-1026

  62. [1998]

    doi: 10.1109/5.726791

    doi

  63. [2016]

    doi: 10.1145/2897937.2898011

  64. [2024]

    doi: 10.1109/ICRC64395.2024.10937023

    2024

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.