Pith. sign in

REVIEW 4 major objections 6 minor 66 references

Sampling from exponential distributions in the time domain with superparamagnetic tunnel junctions

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Measuring the time until a magnetic tunnel junction switches gives one exponential random sample.

desk verdict Solid measurement of exponential first-switch times with a clean circuit; the proposed samplers need a drift bound before their claims hold. read the letter →

arxiv 2412.10317 v2 pith:UTVGWILE submitted 2024-12-13 cs.ET

Temitayo N. Adeyeye , Sidra Gibeault , Daniel P. Lathrop , Matthew W. Daniels , Mark D. Stiles , Jabez J. McClelland , William A. Borders , Jason T. Ryan
show 3 more authors
Philippe Talatchian Ursula Ebels Advait Madhavan
This is my paper · ORCID
classification cs.ET
keywords superparamagnetictunneljunctiontemporalencodingprobabilisticdelaycellexponentialdistributionMetropolis-Hastingsstepperweightedrandomsamplingclockstime-to-digitalconversion
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

Superparamagnetic magnetic tunnel junctions already switch randomly between resistance states, and their dwell times are known to be exponential. This paper turns that known behavior into a usable sampling primitive by measuring, with on-board electronics, the exact time between a current step and the first switching event. The resulting circuit element, called a probabilistic delay cell, produces one tunable exponential random value per switching event, and repeated experiments of about 10,000 events per setting show that the measured times follow the cumulative distribution $F(t) = 1 - e^{-\lambda t}$. The authors argue this matters because exponential variates are the native ingredient of Metropolis-Hastings Monte Carlo and of weighted random sampling, where conventional approaches must approximate an exponential by combining many Bernoulli random bits. If the claim holds, each switching event replaces many bit-draws, potentially saving energy and latency in probabilistic and temporal computing.

What carries the argument

The load-bearing object is the probabilistic delay cell (PDC), a circuit that applies a current step to an SMTJ and converts the first switching time into a digital count. Its statistical identity is the exponential cumulative distribution $F(t) = 1 - e^{-\lambda t}$, with the rate $\lambda$ set by the current amplitude through $\tau_P(I) = \tau_0 \exp[(\Delta E/kT)(1 + I/I_c)^\alpha]$. The companion identity, drawn from the exponential-clocks method, is that when several PDCs run concurrently, the probability that the $j$-th one fires first is $\lambda_j / \sum_i \lambda_i$, which is what turns a race of timers into a weighted random sample. The hardware that carries these identities consists of the transconductance, hysteresis, latching, and counter stages shown in the paper's Fig. 2(a), with a 625 ns path mismatch that produces a known one-bit offset in the counts.

What would settle it

Run a long stationary-current experiment, split the pooled switching times into early and late halves, and compare the two best-fit rates; if the rates differ by more than their statistical uncertainties, the aggregate is a mixture of exponentials rather than one exponential. The paper's Appendix A protocol already provides a template: over 205 s the per-bin mean dwell times spread by 0.042 s against an average within-bin uncertainty of 0.015 s, so the same comparison on the fitted $\lambda$ values would settle whether drift breaks the model.

Watch

Extended reading notes

Core claim

The paper's central claim is that the delay until a superparamagnetic magnetic tunnel junction (SMTJ), initialized in the parallel state, switches to the antiparallel state under a current step is an exponential random variable whose rate $\lambda$ is controlled by the step amplitude. It establishes this by building a probabilistic delay cell: a transconductance stage applies the current step, a programmable hysteresis stage cleans the switching edge, a set-reset latch captures the first rising transition, and a 16-bit counter measures the elapsed time in units of a 500 ns clock. Plots of the cumulative distribution at three current levels (918, 924, and 930 $\mu$A) fit Eq. (1), $F(t) = 1 - e^{-\lambda t}$, with reduced $\chi^2 = 1.92$ for the main dataset, and the mean switching time $\tau = 1/\lambda$ varies exponentially with current as $\tau_P(I) = \tau_0 \exp[(\Delta E/kT)(1 + I/I_c)^\alpha]$, giving nearly two orders of magnitude of tunability. On this basis, the paper proposes two temporal applications: a Metropolis-Hastings stepper in which a probabilistic delay cell racing a deterministic delay cell produces an acceptance bit with probability $e^{-\Theta(\beta\Delta E_i)}$, and a weighted random sampler in which $n$ probabilistic delay cells acting as exponential clocks select index $j$ with probability $\lambda_j / \sum_i \lambda_i$. The applications are proposed and analyzed, not yet built as demonstrated systems.

Load-bearing premise

The claim stands on the switching rate $\lambda$ staying constant while samples are being collected, and the paper's own Appendix A shows the device's mean dwell time drifting over a 205 s run by almost three times the statistical uncertainty, so a long run mixes several exponentials instead of drawing from one.

Editorial extensions

If this is right

  • Each switching event yields one sample from an exponential distribution, so algorithms that need exponential variates can get them without assembling many Bernoulli trials.
  • The rate parameter $\lambda$ is tunable by the applied current, allowing post-fabrication adjustment of the sampling distribution over nearly two orders of magnitude in mean delay.
  • A Metropolis-Hastings acceptance step reduces to comparing the arrival times of a probabilistic and a deterministic delay cell, producing the acceptance probability $e^{-\Theta(\beta\Delta E_i)}$.
  • A set of $n$ probabilistic delay cells with an OR gate implements the exponential-clocks method, drawing the index $j$ with probability $\lambda_j / \sum_i \lambda_i$ in one racing step.
  • The timing measurement is quantized by the clock period (500 ns here), so the number of usable bits per switching event depends on how finely the operating window is divided.

Reading between the lines

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

  • Beyond the paper, the same temporal-encoding scheme should transfer to any Poissonian switching device, such as single-photon avalanche diodes or oxide-memristor random telegraph noise sources, which the paper itself mentions; the PDC would then be a generic analog-to-exponential interface rather than an MTJ-specific circuit.
  • Beyond the paper, the measured drift implies that a single long sampling run should be treated as a mixture of exponentials; a direct test is to split a long stationary-current run in half and compare the two fitted rates, and if they differ, circuit-level calibration or magnetic-field stabilization (noted in Appendix A for newer devices) becomes necessary before the Metropolis-Hastings acceptance
  • Beyond the paper, the bit-depth of the temporal sample is set by clock speed and counter width, so increasing the counter resolution or shortening the clock period directly increases random bits per switching event without changing the device, a design lever that does not exist for state-encoded Bernoulli sampling.
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 / 6 minor

Summary. The paper proposes a probabilistic delay cell built from a superparamagnetic magnetic tunnel junction (SMTJ): the device is biased into the parallel state, a current step puts it into the superparamagnetic regime, and the time of the first P-to-AP switching event is measured with on-board counter electronics. The central experimental claim is that these first-switch times are exponentially distributed with a current-tunable rate. After reporting CDF fits (Fig. 1(c), reduced chi-squared 1.92; Fig. 3), the paper documents device drift (Appendix A, Sec. IV) and then proposes temporal-computing circuits for Bernoulli decisions, a Metropolis-Hastings stepper (Eq. (3)), and a weighted random sampler based on the exponential-clocks method (Eq. (4)). The applications are presented as proposals rather than as measured end-to-end demonstrations.

Significance. If the exponential-switching claim and the stationarity issue can be settled, the probabilistic delay cell is a genuinely useful primitive: a single switching event yields an analog exponential sample, potentially avoiding the repeated Bernoulli trials and Bernoulli-factory overhead that state-encoded SMTJ sampling requires. The paper's strengths are its direct measurement of roughly 10,000 switching events with a reported goodness-of-fit, the clear circuit design for time-domain readout, the qualitatively demonstrated current tunability, and a candid discussion of the device's drift and of the non-physical parameters obtained from the tau(I) fit. Those same strengths make the missing quantitative drift analysis conspicuous, because the proposed applications require fixed rates during a draw.

major comments (4)
  1. [Sec. V, Eq. (4); Appendix A; Sec. IV] The weighted random sampler is analyzed under the assumption that each probabilistic delay cell has a fixed rate lambda_j for the duration of a draw. This is contradicted by the drift documented in Appendix A and Sec. IV: the AP-state probability wanders over 205 s with a spread of 0.042 s versus the 0.015 s average statistical uncertainty in the bins, and the tau(I) fit in Fig. 4 has reduced chi-squared 10.68, which the authors attribute to drift. When lambda_j is time-varying, the realized first-arrival probability is E[lambda_j / sum_k lambda_k] over the drift distribution, not lambda_j / sum_k lambda_k, and the bias is first order in the rate variance. The manuscript gives no bound on this bias, no compensation scheme, and no demonstration that the proposed 'n-sided die' tolerates the observed drift. Please add a quantitative drift analysis, or explicitly restrict the application claim to runs in which stationarity is verified.
  2. [Sec. V, Eq. (3); Sec. IV] The Metropolis-Hastings circuit requires the stochastic delay to be exactly exponential with rate W beta. Under the same device drift, the effective rate is random, so the actual acceptance probability becomes an expectation over the time-varying rate distribution rather than exp(-Theta(beta Delta E_i)). Detailed balance against the target Boltzmann distribution is therefore not guaranteed for the circuit as described. The paper should quantify the drift-induced error in the acceptance probability (or simulate the Metropolis-Hastings step under the drift model of Appendix A) and state the maximum drift for which the sampler remains correct.
  3. [Sec. IV, Eq. (2) and Fig. 4] The quantitative tunability needed by Eq. (4) is not established by the tau(I) characterization. The fit to Eq. (2) has reduced chi-squared 10.68, yields non-physical values of tau_0, Delta E, and I_c, and the exponent alpha is not actually determined, as the footnote states. Since the applications set weights by current addressing, the paper needs at least a per-device calibration procedure (for example, measuring tau at the operating current immediately before a sampling run) or an explicit statement of how the weights are to be calibrated without relying on Eq. (2). As written, only qualitative tunability is demonstrated.
  4. [Sec. IV, Fig. 3 inset] The long-time tail is part of the exponential claim, but the inset of Fig. 3 shows a deviation from the fit that the text attributes to 'device variability' without quantification. The applications compare exponential samples to deterministic thresholds and to one another, so tail behavior matters. Please quantify the long-time deviation (for example, with a two-component fit or a tail-specific test) and state how the proposed circuits tolerate it.
minor comments (6)
  1. [Introduction and Sec. IV] There are typographical issues: 'Kramer's' should be 'Kramers' in Sec. IV, and 'come existing methods' in the Introduction should be 'some existing methods.'
  2. [Fig. 4] The axis labels of Fig. 4 appear in the manuscript as corrupted glyph sequences ('/uni000000...'); please ensure the final figure is legible.
  3. [Fig. 3] Report reduced chi-squared or another goodness-of-fit statistic for the three CDF fits in Fig. 3, as is done for Fig. 1(c), rather than relying on visual inspection.
  4. [Sec. III A] The residual 625 ns timing difference between the signal and reference paths is said to cause a systematic one-bit offset; specify whether the reported time values have been corrected for this offset.
  5. [Sec. IV] The Fokker-Planck/macrospin analysis mentioned in Sec. IV is not presented. Either include the governing equation and representative results, or cite the source for the statement that short-time corrections become unimportant for Delta E/kT > 15.
  6. [Sec. VI] The energy and throughput estimates in Sec. VI depend on assumed device parameters such as 1 V, 10 microampere, 100 ns, and eight time bins; label these explicitly as estimates and provide a model or reference for the per-bit energy comparison.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the exponential distribution is measured directly, and the application circuits follow from standard exponential-clocks algebra.

full rationale

The paper's central exponential claim is an experimental characterization, not a derivation from a fitted target. Eq. (1) is the standard exponential CDF assumed from Kramer's escape/Poisson switching and then tested against measured first-switching times; Fig. 1(c) reports a reduced chi-squared of 1.92 for that fit. The applications in Sec. V are algebraic consequences of that CDF and the standard exponential-clocks theorem (Eq. 4); they do not restate a fitted parameter under a new name. Eq. (2) is explicitly an empirical fit whose extracted parameters the authors say are not physically meaningful, and using it to argue for current-tunable rates and Boltzmann-weight selection is a design consequence, not a prediction validated against the same data. Self-citations (e.g., Refs. [3], [14], and temporal-computing references) supply background and prior context; the present paper independently measures both the exponential distribution and the device drift, so these citations are not load-bearing. The acknowledged nonstationarity and drift in Appendix A and Sec. IV is a validity and robustness concern for long sampling runs, but it is not circularity because the paper does not define the exponential rate in terms of the Metropolis-Hastings or weighted-sampler outputs. No step reduces by construction to its own inputs.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The paper's core contribution is an experimental characterization of known SMTJ physics, so the load-bearing assumptions are about the validity of the single-rate exponential model and the stability of the device. The only fitted parameters in a model equation are in Eq. (2), which the authors themselves describe as physically meaningless.

free parameters (2)
  • Parameters tau0, Delta E, Ic in Eq. (2) = Not reported; stated as non-physical
    Fitted to the tau vs I data in Fig. 4; the paper states the extracted values are not physically meaningful.
  • Exponent alpha in Eq. (2) = 1
    Assumed alpha = 1 in the model; the paper says results are consistent with alpha = 1 but not inconsistent with other values.
assumptions (4)
  • domain assumption SMTJ switching is a two-state Markov process with a single constant rate lambda (Kramers escape / Poisson process).
    Introduced in Sec. IV to justify Eq. (1). The paper's macrospin analysis shows short-time corrections for low barriers, so this is approximate.
  • domain assumption Device properties are stationary over the measurement window.
    Required for the exponential distribution to have a fixed rate. Appendix A shows drift over 205 s, violating this on long timescales.
  • domain assumption Macrospin Fokker-Planck model captures the relevant switching dynamics.
    Used in Sec. IV to estimate deviations from exponentiality; micromagnetic effects are noted as not captured.
  • domain assumption Constant temperature T = 300 K.
    Assumed in Eq. (2) and the energy-barrier discussion; temperature drift is proposed as a cause of device drift.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Sampling from exponential distributions in the time domain with superparamagnetic tunnel junctions." pith.science (2026). https://pith.science/paper/UTVGWILE

@misc{pith2026241210317,
  author       = {Pith},
  title        = {Pith review of: Sampling from exponential distributions in the time domain with superparamagnetic tunnel junctions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UTVGWILE}},
  note         = {Machine review of arXiv:2412.10317}
}
read the original abstract

In the superparamagnetic regime, magnetic tunnel junctions switch between two resistance states due to random thermal fluctuations. The dwell time distribution in each state is exponential. We sample this distribution using a temporal encoding scheme, in which information is encoded in the time at which the device switches between its resistance states. We then develop a circuit element known as a probabilistic delay cell that applies an electrical current step to a superparamagnetic tunnel junction and a temporal measurement circuit that measures the timing of the first switching event. Repeated experiments confirm that these times are exponentially distributed. Temporal processing methods then allow us to digitally compute with these exponentially distributed probabilistic delay cells. We describe how to use these circuits in a Metropolis-Hastings stepper and in a weighted random sampler, both of which are computationally intensive applications that benefit from the efficient generation of exponentially distributed random numbers.

Figures

Figures reproduced from arXiv: 2412.10317 by the authors.

Figure 1
Figure 1. FIG. 1. Extracting temporal stochasticity with a current step: [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Onboard timing measurement of a probabilistic delay cell: Panel (a) shows the timing measurement circuit with the [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. Cumulative distribution function (CDF) of switching [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Exponential dependence of mean switching times [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5. Probabilistic sampling with temporal circuits: Panel [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6 [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 58 canonical work pages

  1. [1]

    exponential clocks

    Panel (a) also shows a single transistor implementation of the inhibit gate from [44]. Panel (b) shows the simplest version of a circuit that can make Boolean decisions based on a temporal sample from an exponential distribution. A deterministic delay cell (DDC) and a probabilistic delay cell (PDC) are fed into the B (for blocking) and I (for inhibited) i...

    work page 2000

  2. [2]

    T ransconductance Stage This section describes the equation for the transcon- ductance stage in Fig. 2(a). The transconductance stage acts as a tunable current source for the SMTJ by us- ing the high gain of the operational amplifier to pin the voltages across the resistor RT C. This causes the cur- rent through the device to be solely determined by the r...

    work page

  3. [3]

    Hysteresis State This section describes the equation governing the hys- teresis stage presented in Fig. 2(a). V + is the voltage /uni00000013/uni00000015/uni00000018/uni00000018/uni00000013/uni0000001a/uni00000018/uni00000014/uni00000013/uni00000013/uni00000014/uni00000015/uni00000018/uni00000014/uni00000018/uni00000013/uni00000014/uni0000001a/uni00000018...

    work page 2000

  4. [4]

    Vodenicarevic, N

    D. Vodenicarevic, N. Locatelli, A. Mizrahi, J. S. Fried- man, A. F. Vincent, M. Romera, A. Fukushima, K. Yakushiji, H. Kubota, S. Yuasa, S. Tiwari, J. Grol- lier, and D. Querlioz, Low-energy truly random number generation with superparamagnetic tunnel junctions for unconventional computing, Physical Review Applied 8, 054045 (2017)

    work page 2017

  5. [5]

    Borders, H

    W. Borders, H. Ohno, and S. Fukami, Nanosecond ran- dom telegraph noise in in-plane magnetic tunnel junc- tions, Physical review letters 126, 117202 (2021)

    work page 2021

  6. [6]

    Talatchian, M

    P. Talatchian, M. W. Daniels, A. Madhavan, M. R. Pu- fall, E. Ju´ e, W. H. Rippard, J. J. McClelland, and M. D. Stiles, Mutual control of stochastic switching for two electrically coupled superparamagnetic tunnel junctions, Physical Review B 104, 054427 (2021)

    2021

  7. [7]

    Niazi, S

    S. Niazi, S. Chowdhury, N. A. Aadit, M. Mohseni, Y. Qin, and K. Y. Camsari, Training deep Boltzmann networks with sparse Ising machines, Nature Electronics , 1 (2024)

    work page 2024

  8. [8]

    W. A. Borders, A. Z. Pervaiz, S. Fukami, K. Y. Cam- sari, H. Ohno, and S. Datta, Integer factorization using stochastic magnetic tunnel junctions, Nature 573, 390 (2019)

    2019

Show all 66 references
  1. [9]

    Kittel and H

    C. Kittel and H. Kroemer, Thermal physics (Macmillan, 1980)

    1980

  2. [10]

    Kirkpatrick, C

    S. Kirkpatrick, C. D. Gelatt Jr, and M. P. Vecchi, Opti- mization by simulated annealing, science220, 671 (1983)

    1983

  3. [11]

    H¨ ubschle-Schneider and P

    L. H¨ ubschle-Schneider and P. Sanders, Parallel weighted random sampling, ACM Transactions on Mathematical Software (TOMS) 48, 1 (2022)

    2022

  4. [12]

    Chowdhury, A

    S. Chowdhury, A. Grimaldi, N. A. Aadit, S. Niazi, M. Mohseni, S. Kanai, H. Ohno, S. Fukami, L. Theogara- jan, G. Finocchio, et al., A full-stack view of probabilistic computing with p-bits: devices, architectures, and algo- rithms, IEEE Journal on Exploratory Solid-State Com- ...

    2023

  5. [13]

    Misra, L

    S. Misra, L. C. Bland, S. G. Cardwell, J. A. C. In- corvia, C. D. James, A. D. Kent, C. D. Schuman, J. D. Smith, and J. B. Aimone, Probabilistic neural computing with stochastic devices, Advanced Materials 35, 2204569 (2023)

    2023

  6. [14]

    K. Y. Camsari, Probabilistic computing with p-bits: Op- timization, machine learning and quantum simulation, in 2024 IEEE International Magnetic Conference-Short pa- pers (INTERMAG Short papers)(IEEE, 2024) pp. 1–2

    2024

  7. [15]

    B. A. Cipra, An introduction to the Ising model, The American Mathematical Monthly 94, 937 (1987)

    1987

  8. [16]

    K. Y. Camsari, R. Faria, B. M. Sutton, and S. Datta, Stochastic p-bits for invertible logic, Physical Review X 7, 031014 (2017)

    2017

  9. [17]

    Gibeault, T

    S. Gibeault, T. N. Adeyeye, L. A. Pocher, D. P. Lath- rop, M. W. Daniels, M. D. Stiles, J. J. McClelland, W. A. Borders, J. T. Ryan, P. Talatchian,et al., Programmable electrical coupling between stochastic magnetic tunnel junctions, Physical Review Applied 21, 034064 (2024)

    2024

  10. [18]

    N. A. Aadit, A. Grimaldi, M. Carpentieri, L. Theogara- jan, J. M. Martinis, G. Finocchio, and K. Y. Camsari, Massively parallel probabilistic computing with sparse Ising machines, Nature Electronics 5, 460 (2022)

    2022

  11. [19]

    N. S. Singh, K. Kobayashi, Q. Cao, K. Selcuk, T. Hu, S. Niazi, N. A. Aadit, S. Kanai, H. Ohno, S. Fukami, et al., CMOS plus stochastic nanomagnets enabling het- erogeneous computers for probabilistic inference and learning, Nature Communications 15, 2685 (2024)

    2024

  12. [20]

    J. Si, S. Yang, Y. Cen, J. Chen, Z. Yao, D.-J. Kim, K. Cai, J. Yoo, X. Fong, and H. Yang, Energy- efficient superparamagnetic Ising machine and its appli- cation to traveling salesman problems, arXiv preprint arXiv:2306.11572 (2023)

    2023 arXiv

  13. [21]

    Hespanha and K

    J. Hespanha and K. C ¸ amsari, Markov chain Monte Carlo for koopman-based optimal control, IEEE Control Sys- tems Letters (2024)

    2024

  14. [22]

    Lucas, Ising formulations of many np problems, Fron- tiers in Physics 2, 5 (2014)

    A. Lucas, Ising formulations of many np problems, Fron- tiers in Physics 2, 5 (2014)

    2014

  15. [23]

    Hamburg, P

    M. Hamburg, P. Kocher, and M. E. Marson, Anal- ysis of intel’s ivy bridge digital random number generator, Online: http://www. cryptography. com/public/pdf/Intel TRN G Report 20120312. pdf (2012)

    2012

  16. [24]

    1/DeveloperGuide/text/SD/Security/OpTee.html? #hwkey-agent-trusted-application (2024), accessed: 2024-11-12

    NVIDIA Hardware TRNG trusted application api, https://docs.nvidia.com/jetson/archives/r35. 1/DeveloperGuide/text/SD/Security/OpTee.html? #hwkey-agent-trusted-application (2024), accessed: 2024-11-12

    2024

  17. [25]

    W. Wan, R. Kubendran, C. Schaefer, S. B. Eryilmaz, W. Zhang, D. Wu, S. Deiss, P. Raina, H. Qian, B. Gao, et al., A compute-in-memory chip based on resistive random-access memory, Nature 608, 504 (2022)

    2022

  18. [26]

    N. A. Aadit, A. Grimaldi, M. Carpentieri, L. Theogara- jan, G. Finocchio, and K. Y. Camsari, Computing with invertible logic: Combinatorial optimization with proba- bilistic bits, in 2021 IEEE International Electron Devices Meeting (IEDM) (IEEE, 2021) pp. 40–3

    2021

  19. [27]

    M. D. Stiles and A. Zangwill, Anatomy of spin-transfer torque, Physical Review B 66, 014407 (2002)

    2002

  20. [28]

    K. Lee, J. Bak, Y. Kim, C. Kim, A. Antonyan, D. Chang, S. Hwang, G. Lee, N. Ji, W. Kim, et al., 1gbit high density embedded STT-MRAM in 28nm fdsoi technol- ogy, in 2019 IEEE International Electron Devices Meet- ing (IEDM) (IEEE, 2019) pp. 2–2

    2019

  21. [29]

    D. C. Worledge, Spin-transfer-torque MRAM: the next revolution in memory, in 2022 IEEE international mem- ory workshop (IMW)(IEEE, 2022) pp. 1–4

    2022

  22. [30]

    H. Zhao, B. Glass, P. K. Amiri, A. Lyle, Y. Zhang, Y.- J. Chen, G. Rowlands, P. Upadhyaya, Z. Zeng, J. Ka- tine, et al., Sub-200 ps spin transfer torque switching in in-plane magnetic tunnel junctions with interface perpen- dicular anisotropy, Journal of Physics D: Applied Phy...

    2011

  23. [31]

    Schnitzspan, M

    L. Schnitzspan, M. Kl¨ aui, and G. Jakob, Nanosecond true random number generation with superparamagnetic tunnel junctions–identification of joule heating and spin- transfer-torque effects, arXiv preprint arXiv:2301.05694 13 (2023)

    2023 arXiv

  24. [32]

    Fukushima, T

    A. Fukushima, T. Seki, K. Yakushiji, H. Kubota, H. Ima- mura, S. Yuasa, and K. Ando, Spin dice: A scalable truly random number generator based on spintronics, Applied Physics Express 7, 083001 (2014)

    2014

  25. [33]

    Shukla, L

    A. Shukla, L. Heller, M. G. Morshed, L. Rehm, A. W. Ghosh, A. D. Kent, and S. Rakheja, A true random num- ber generator for probabilistic computing using stochas- tic magnetic actuated random transducer devices, in2023 24th International Symposium on Quality Electronic De- sig...

    2023

  26. [34]

    Von Neumann et al., Various techniques used in con- nection with random digits, John von Neumann, Col- lected Works 5, 768 (1963)

    J. Von Neumann et al., Various techniques used in con- nection with random digits, John von Neumann, Col- lected Works 5, 768 (1963)

    1963

  27. [35]

    Keane and G

    M. Keane and G. L. O’Brien, A bernoulli factory, ACM Transactions on Modeling and Computer Simulation (TOMACS) 4, 213 (1994)

    1994

  28. [36]

    Dughmi, J

    S. Dughmi, J. Hartline, R. D. Kleinberg, and R. Ni- azadeh, Bernoulli factories and black-box reductions in mechanism design, Journal of the ACM (JACM) 68, 1 (2021)

    2021

  29. [37]

    Soumah, L

    L. Soumah, L. Desplat, N.-T. Phan, A. S. E. Valli, A. Madhavan, F. Disdier, S. Auffret, R. Sousa, U. Ebels, and P. Talatchian, Nanosecond stochastic operation in perpendicular superparamagnetic tunnel junctions, arXiv preprint arXiv:2402.03452 (2024)

    2024 arXiv

  30. [38]

    W. H. Butler, T. Mewes, C. K. Mewes, P. Visscher, W. H. Rippard, S. E. Russek, and R. Heindl, Switching distri- butions for perpendicular spin-torque devices within the macrospin approximation, IEEE Transactions on Mag- netics 48, 4684 (2012)

    2012

  31. [39]

    B. C. McGoldrick and J. Z. Sun, Settling time of current- tunable probabilistic bit’s distribution, IEEE Magnetics Letters 13, 1 (2022)

    2022

  32. [40]

    While our results are consistent with α = 1, they are not inconsistent with other values because the frac- tional variation of the current in our measurements is so small

    Different models predict different values for the exponent α [63]. While our results are consistent with α = 1, they are not inconsistent with other values because the frac- tional variation of the current in our measurements is so small. Taking any value for 1 ≤ α ≤ 2 does no...

  33. [41]

    D. B. Thomas, L. Howes, and W. Luk, A comparison of cpus, gpus, fpgas, and massively parallel processor ar- rays for random number generation, in Proceedings of the ACM/SIGDA international symposium on Field pro- grammable gate arrays(2009) pp. 63–72

    2009

  34. [42]

    Jun and P

    B. Jun and P. Kocher, The Intel random number gener- ator, Tech. Rep. (Intel Corporation, 1999)

    1999

  35. [43]

    A. J. Walker, An efficient method for generating dis- crete random variables with general distributions, ACM Transactions on Mathematical Software (TOMS) 3, 253 (1977)

    1977

  36. [44]

    P. Wang, C. Li, J. Wang, T. Wang, L. Zhang, J. Leng, Q. Chen, and M. Guo, Skywalker: Efficient alias-method- based graph sampling and random walk on gpus, in 2021 30th International Conference on Parallel Architectures and Compilation Techniques (PACT)(IEEE, 2021) pp. 304–317

    2021

  37. [45]

    D. B. Thomas and W. Luk, Sampling from the expo- nential distribution using independent bernoulli variates, in 2008 International Conference on Field Programmable Logic and Applications(IEEE, 2008) pp. 239–244

    2008

  38. [46]

    Madhavan, T

    A. Madhavan, T. Sherwood, and D. Strukov, Race logic: a hardware acceleration for dynamic programming algo- rithms, SIGARCH Comput. Archit. News 42, 517–528 (2014)

    2014

  39. [47]

    Tzimpragos, A

    G. Tzimpragos, A. Madhavan, D. Vasudevan, D. Strukov, and T. Sherwood, Boosted race trees for low energy clas- sification, in Proceedings of the Twenty-Fourth Interna- tional Conference on Architectural Support for Program- ming Languages and Operating Systems(2019) pp. 215– 228

    2019

  40. [48]

    Chen and J

    Z. Chen and J. Gu, A time-domain computing acceler- ated image recognition processor with efficient time en- coding and non-linear logic operation, IEEE Journal of Solid-State Circuits 54, 3226 (2019)

    2019

  41. [49]

    Sayal, S

    A. Sayal, S. T. Nibhanupudi, S. Fathima, and J. P. Kulka- rni, A 12.08-tops/w all-digital time-domain cnn engine using bi-directional memory delay lines for energy effi- cient edge computing, IEEE Journal of Solid-State Cir- cuits 55, 60 (2019)

    2019

  42. [50]

    Madhavan, T

    A. Madhavan, T. Sherwood, and D. Strukov, A 4-mm 2 180-nm-CMOS 15-giga-cell-updates-per-second dna se- quence alignment engine based on asynchronous race con- ditions, in 2017 IEEE Custom Integrated Circuits Con- ference (CICC)(IEEE, 2017) pp. 1–4

    2017

  43. [51]

    K. Kim, W. Yu, and S. Cho, A 9 bit, 1.12 ps resolution 2.5 b/stage pipelined time-to-digital converter in 65 nm CMOS using time-register, IEEE Journal of Solid-State Circuits 49, 1007 (2014)

    2014

  44. [52]

    Byrd and Z

    J. Byrd and Z. Lipton, What is the effect of importance weighting in deep learning?, in International conference on machine learning(PMLR, 2019) pp. 872–881

    2019

  45. [53]

    Kurant, M

    M. Kurant, M. Gjoka, C. T. Butts, and A. Markopoulou, Walking on a graph with a magnifying glass: strati- fied sampling via weighted random walks, in Proceedings of the ACM SIGMETRICS joint international confer- ence on Measurement and modeling of computer systems (2011) pp. 281–292

    2011

  46. [54]

    Kalal, J

    Z. Kalal, J. Matas, and K. Mikolajczyk, Weighted sam- pling for large-scale boosting., in BMVC, Vol. 2 (2008) p. 5

    2008

  47. [55]

    Arratia, On the amount of dependence in the prime factorization of a uniform random integer, arXiv preprint arXiv:1305.0941 (2013)

    R. Arratia, On the amount of dependence in the prime factorization of a uniform random integer, arXiv preprint arXiv:1305.0941 (2013)

    2013 arXiv

  48. [56]

    Madhavan, M

    A. Madhavan, M. W. Daniels, and M. D. Stiles, Temporal state machines: Using temporal memory to stitch time- based graph computations, J. Emerg. Technol. Comput. Syst. 17, 10.1145/3451214 (2021)

    2021 doi

  49. [57]

    Gretsch, P

    R. Gretsch, P. Song, A. Madhavan, J. Lau, and T. Sher- wood, Energy efficient convolutions with temporal arith- metic, in Proceedings of the 29th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 2, ASPLOS ’24 (Associ...

    2024

  50. [58]

    H. Nair, J. P. Shen, and J. E. Smith, A microarchitecture implementation framework for online learning with tem- poral neural networks, in 2021 IEEE Computer Society Annual Symposium on VLSI (ISVLSI)(IEEE, 2021) pp. 266–271

    2021

  51. [59]

    Kaiser, W

    J. Kaiser, W. A. Borders, K. Y. Camsari, S. Fukami, H. Ohno, and S. Datta, Hardware-aware in situ learning based on stochastic magnetic tunnel junctions, Physical Review Applied 17, 014016 (2022)

    2022

  52. [60]

    Whitehead, Z

    W. Whitehead, Z. Nelson, K. Y. Camsari, and L. Theog- arajan, Cmos-compatible ising and potts annealing us- ing single-photon avalanche diodes, Nature Electronics 6, 1009 (2023). 14

    2023

  53. [61]

    Stanco, D

    A. Stanco, D. G. Marangon, G. Vallone, S. Burri, E. Charbon, and P. Villoresi, Efficient random number generation techniques for cmos single-photon avalanche diode array exploiting fast time tagging units, Physical Review Research 2, 023287 (2020)

    2020

  54. [62]

    S. Choi, Y. Yang, and W. Lu, Random telegraph noise and resistance switching analysis of oxide based resistive memory, Nanoscale 6, 400 (2014)

    2014

  55. [63]

    X. Li, T. Zanotti, T. Wang, K. Zhu, F. M. Puglisi, and M. Lanza, Random telegraph noise in metal-oxide mem- ristors for true random number generators: A materi- als study, Advanced Functional Materials 31, 2102172 (2021)

    2021

  56. [64]

    Narayanan, B

    V. Narayanan, B. M. Fleischer, and B. A. Chap- pell, Static timing analysis for self resetting circuits, in Computer-Aided Design, International Conference on (IEEE Computer Society, 1996) pp. 119–119

    1996

  57. [65]

    Filanovsky and H

    I. Filanovsky and H. Baltes, CMOS schmitt trigger de- sign, IEEE Transactions on Circuits and Systems I: Fun- damental Theory and Applications 41, 46 (1994)

    1994

  58. [66]

    J. Z. Sun and C. Safranski, Metrology and metrics for spin-transfer-torque switched magnetic tunnel junctions in memory applications, Journal of Magnetism and Mag- netic Materials 563, 169878 (2022)

    2022

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.