pith. sign in

arxiv: 2606.13179 · v1 · pith:2V66543Inew · submitted 2026-06-11 · 💻 cs.ET · cs.AI· cs.AR· cs.NE

Modern analog computing for solving differential and matrix equations

Zhong Sun , Piergiulio Mannocci , Manuel Le Gallo , Abu Sebastian This is my paper

Pith reviewed 2026-06-27 05:11 UTC · model grok-4.3

classification 💻 cs.ET cs.AIcs.ARcs.NE
keywords analog computingresistive memorydifferential equationsmatrix equationsin-memory computinghardware acceleratorsscientific computing
0
0 comments X

The pith

Resistive memory arrays implement analog operators that solve differential and matrix equations efficiently.

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

The paper surveys modern analog computing for solving differential equations and matrix equations, tasks central to scientific computing and artificial intelligence. It defines three core primitives—differential equation solving, matrix equation solving, and matrix-vector multiplication—and traces the connections among them. Hardware options ranging from discrete components and analog CMOS circuits to resistive memory devices are examined, with resistive memory arrays singled out for their implementation efficiency. Recent demonstrations using these arrays and advanced circuits are reviewed, followed by discussion of applications, precision and scalability limitations, potential fixes, links to in-memory computing, and the distinctive computational complexity of analog approaches. The work assembles these elements into one perspective that positions analog computing as a route to next-generation performance.

Core claim

Modern analog computing, realized through resistive memory arrays and analog CMOS circuits, supplies hardware primitives that directly perform the matrix-vector operations underlying the solution of differential and matrix equations, thereby connecting these tasks and delivering efficiency advantages for data-intensive workloads.

What carries the argument

Resistive memory arrays configured to execute analog matrix-vector multiplications that solve differential and matrix equations.

If this is right

  • Analog implementations can reduce energy and latency for the matrix operations that dominate scientific and AI workloads.
  • Resistive memory arrays enable in-memory computation that avoids repeated data movement between memory and processor.
  • Progress in both analog CMOS and resistive technologies supports concrete circuit designs for equation solvers.
  • The same hardware primitives can address multiple equation types through shared matrix-vector operations.

Where Pith is reading between the lines

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

  • Hybrid analog-digital systems could use resistive arrays for the bulk linear algebra while reserving digital logic for nonlinear corrections and high-precision refinement.
  • If array sizes scale, the same platforms might accelerate large-scale simulations in climate or fluid dynamics that currently rely on iterative matrix solvers.
  • The unique complexity of analog computation may require new algorithms that tolerate bounded noise rather than exact arithmetic.

Load-bearing premise

Precision, scalability, and integration problems in analog hardware can be resolved sufficiently for the technology to move beyond laboratory demonstrations.

What would settle it

An experiment showing that resistive memory arrays cannot reach the accuracy or array size needed to solve representative systems of differential or matrix equations with error below a practical threshold would disprove the efficiency claim.

Figures

Figures reproduced from arXiv: 2606.13179 by Abu Sebastian, Manuel Le Gallo, Piergiulio Mannocci, Zhong Sun.

Figure 1
Figure 1. Figure 1: Analog computing primitives, including solving ordinary differential equation, solving matrix equation, and performing matrix multiplication, each of which are all is implementable with analog hardware. The algorithmic approach in digital computing is also illustrated for comparison. Notably, algorithms are partially involved in analog computing as well, to decompose the computing tasks. For example, a dif… view at source ↗
Figure 2
Figure 2. Figure 2: Circuit schematics of the three computing primitives. a, Basic analog operators. Circuit schematics for b, matrix-vector multiplication, c, solving an ordinary differential equation, and d, solving a matrix equation. In these schematics, the constant input operators have been omitted for clarity and instead are represented by a floating arrow [PITH_FULL_IMAGE:figures/full_fig_p034_2.png] view at source ↗
Figure 5
Figure 5. Figure 5: Resistive memory circuits for solving matrix equations. a [PITH_FULL_IMAGE:figures/full_fig_p035_5.png] view at source ↗
Figure 1
Figure 1. Figure 1: Analog computer Partial differential equation Ordinary differential equation Matrix equation Matrix multiplication Digital computer Space discretization Time discretization Algorithm (Iterative) Program Hardware mapping [PITH_FULL_IMAGE:figures/full_fig_p037_1.png] view at source ↗
read the original abstract

In recent years, driven by the computational demands of data-intensive applications such as artificial intelligence and scientific computing, analog computing has gained renewed interest. Given the diversity of computational tasks and recent advancements in analog CMOS circuits and resistive memory technologies, we refer to the evolving landscape as modern analog computing. In this context, we identify three core computational primitives: solving differential equations, solving matrix equations, and performing matrix-vector multiplications, and we explore the connections among them. We also examine various hardware implementations of these analog computing operators, including those built with discrete components, integrated circuits, and resistive memory devices. Among these, resistive memory arrays emerge as particularly promising due to their implementation efficiency. The paper then surveys recent progress in leveraging modern analog computing to solve differential and matrix equations using both advanced analog CMOS circuits and resistive memory arrays. Finally, we discuss the applications of these circuits, the precision and scalability issues and their potential solutions, the relationship with in-memory computing, and the unique computational complexity of analog computing. This paper provides a unified perspective on analog computing, highlighting its strengths, current developments, and challenges, and positioning it as a pivotal enabler of next-generation computational frontiers.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

0 major / 2 minor

Summary. The manuscript is a survey on modern analog computing for solving differential and matrix equations. It identifies three core primitives (solving differential equations, solving matrix equations, and matrix-vector multiplications) and their connections, reviews hardware options including discrete components, integrated circuits, and resistive memory devices (highlighting the latter for implementation efficiency), surveys recent progress with analog CMOS circuits and resistive arrays, and discusses applications, precision/scalability issues with potential solutions, links to in-memory computing, and analog computational complexity. The goal is a unified perspective positioning analog computing for next-generation applications.

Significance. As a survey synthesizing existing literature, the paper could serve as a useful reference by linking primitives across hardware platforms and noting the efficiency of resistive memory arrays based on prior work. It addresses the stress-test concern on challenges by discussing potential solutions in the relevant section, so that assumption does not undermine the manuscript. No new quantitative benchmarks, machine-checked proofs, or reproducible code are provided (appropriate for a survey), and the assessment of solutions remains qualitative rather than data-driven.

minor comments (2)
  1. [Introduction] The abstract introduces 'modern analog computing' as a term for the evolving landscape; the introduction should explicitly state whether this is a coined phrase or drawn from prior literature, with supporting citations.
  2. [Discussion of precision and scalability issues] The section on precision and scalability issues lists potential solutions at a high level; adding one or two concrete references to experimental demonstrations of those solutions would improve clarity without altering the survey nature.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary, significance assessment, and recommendation of minor revision. The manuscript is a survey paper, and we appreciate the recognition that it synthesizes literature without requiring new benchmarks or proofs. No specific major comments were provided in the report, so we have no point-by-point responses to address. We will handle any minor issues during revision.

Circularity Check

0 steps flagged

Survey paper with no internal derivations or predictions

full rationale

The manuscript is a survey that identifies computational primitives, examines hardware options including resistive memory arrays, and reviews progress from external literature. No equations, predictions, or derivations are presented that could reduce to fitted parameters, self-definitions, or self-citation chains. All claims are positioned as syntheses of cited prior work, making the paper self-contained against external benchmarks with no circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is a review paper; the central perspective rests on the accuracy of the surveyed literature rather than new derivations. No free parameters, axioms, or invented entities are introduced.

pith-pipeline@v0.9.1-grok · 5743 in / 990 out tokens · 19633 ms · 2026-06-27T05:11:14.772199+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

125 extracted references · 5 canonical work pages · 1 internal anchor

  1. [1]

    Analog VLSI and neural systems (Addison-Wesley, 1989)

    Carver, M. Analog VLSI and neural systems (Addison-Wesley, 1989)

    1989

  2. [2]

    Patterson, D. et al. Carbon emissions and large neural network training. arXiv preprint arXiv:2104.10350 (2021)

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  3. [3]

    Where are we going, Phil Morse? Changing agendas and the rhetoric of obviousness in the transformation of computing at MIT, 1939-1957

    Owens, L. Where are we going, Phil Morse? Changing agendas and the rhetoric of obviousness in the transformation of computing at MIT, 1939-1957. IEEE Ann. Hist. Comput. 18, 34–41 (1996)

    1939

  4. [4]

    Graca, D. S. Some recent developments on Shannon’s GPAC. Math. Log. Quart. 50, 473–485 (2004)

    2004

  5. [5]

    Freeth, T. et al. A Model of the Cosmos in the ancient Greek Antikythera Mechanism. Sci. Rep. 11, 5821 (2021)

    2021

  6. [6]

    Freeth, T. et al. Decoding the ancient Greek astronomical calculator known as the Antikythera Mechanism. Nature 444, 587–591 (2006)

    2006

  7. [7]

    Mallock, R. R. M. An electrical calculating machine. Proc. R. Soc. Lond. A 140, 457–483 (1933)

    1933

  8. [8]

    Hartree, D. R. The Bush Differential Analyser and its Applications. Nature 146, 319–323 (1940)

    1940

  9. [9]

    Not your father’s analog computer

    Tsividis, Y . Not your father’s analog computer. IEEE Spectr. 55, 38–43 (2017)

    2017

  10. [10]

    Haensch, W. et al. The next generation of deep learning hardware: analog computing. Proc. IEEE 107, 108–122 (2019)

    2019

  11. [11]

    Sun, Z. et al. Invited tutorial: Analog matrix computing with crosspoint resistive memory arrays. IEEE Trans. Circuits Syst. II Exp. Briefs 69, 3024–3029 (2022)

    2022

  12. [12]

    Care, C. P. From analogy-making to modelling: the history of analog computing as a modelling technology. PhD thesis, University of Warwick (2008)

    2008

  13. [13]

    Sun, Z. et al. A full spectrum of computing-in-memory technologies. Nat. Electron. 6, 823–835 (2023)

    2023

  14. [14]

    Analog computation (McGraw-Hill, 1960)

    Jackson, A. Analog computation (McGraw-Hill, 1960)

    1960

  15. [15]

    Korn, G. et al. Electronic Analog and Hybrid Computers (McGraw-Hill, 1972)

    1972

  16. [16]

    Analog Computing (Oldenbourg Wissenschaftsverlag, 2013)

    Ulmann, B. Analog Computing (Oldenbourg Wissenschaftsverlag, 2013)

    2013

  17. [17]

    George, S. et al. A programmable and configurable mixed-mode FPAA SoC. IEEE Trans. V ery Large Scale Integr. VLSI Syst. 24, 2253–2261 (2016)

    2016

  18. [18]

    Large-scale field-programmable analog arrays

    Hasler, J. Large-scale field-programmable analog arrays. Proc. IEEE 108, 1283–1302 (2019)

    2019

  19. [19]

    Sun, Z. et al. Time complexity of in-memory matrix-vector multiplication. IEEE Trans. Circuits Syst. II Exp. Briefs 68, 2785–2789 (2021)

    2021

  20. [20]

    Le Gallo, M. et al. Mixed-precision in-memory computing. Nat. Electron. 1, 246–253 (2018)

    2018

  21. [21]

    Zidan, M. A. et al. A general memristor-based partial differential equation solver. Nat. Electron. 1, 411–420 (2018)

    2018

  22. [22]

    Sun, Z. et al. Solving matrix equations in one step with cross-point resistive arrays. Proc. Natl Acad. Sci. USA 116, 4123–4128 (2019)

    2019

  23. [23]

    Huang, Y . et al. Hybrid analog-digital solution of nonlinear partial differential equations. in Proc. IEEE/ACM Int. Symp. Microarchit. 665–678 (2017)

    2017

  24. [24]

    Golub, G. H. et al. Matrix Computations (Johns Hopkins University Press, 2013)

    2013

  25. [25]

    Shannon, C. E. Mathematical theory of the differential analyzer. J. Math. Phys. 20, 337– 354 (1941)

    1941

  26. [26]

    Bournez, O. et al. The general purpose analog computer and computable analysis are two equivalent paradigms of analog computation. in Theory and Applications of Models of Computation. 631–643 (Springer, 2006)

    2006

  27. [27]

    Milos, N. et al. Analog computer on a chip – compiling solutions. in Embed. World Conf. 1–8 (2020)

    2020

  28. [28]

    Cowan, G. et al. A VLSI analog computer/digital computer accelerator. IEEE J. Solid- State Circuits 41, 42–53 (2006)

    2006

  29. [29]

    Guo, N. et al. Continuous-time hybrid computation with programmable nonlinearities. in Eur. Solid-State Circuits Conf. 279–282 (2015)

    2015

  30. [30]

    Huang, Y . et al. Evaluation of an analog accelerator for linear algebra. SIGARCH Comput. Archit. News 44, 570–582 (2016)

    2016

  31. [31]

    Berry, C. E. et al. A Computer for Solving Linear Simultaneous Equations. J. App. Phys. 17, 262 (1946)

    1946

  32. [32]

    Analog and Hybrid Computer Programming (De Gruyter, 2020)

    Ulmann, B. Analog and Hybrid Computer Programming (De Gruyter, 2020)

    2020

  33. [33]

    A precise four-quadrant multiplier with subnanosecond response

    Gilbert, B. A precise four-quadrant multiplier with subnanosecond response. IEEE J. Solid-State Circuits 3, 365–373 (1968)

    1968

  34. [34]

    Maryan, M. M. et al. A MOS translinear cell-based configurable block for current-mode analog signal processing. Analog Integr. Circuits Signal Process. 92, 1–13 (2017)

    2017

  35. [35]

    Kub, F. J. et al. Programmable analog vector-matrix multipliers. IEEE J. Solid-State Circuits 25, 207–214 (1990)

    1990

  36. [36]

    Borghetti, J. et al. ‘Memristive’ switches enable ‘stateful’ logic operations via material implication. Nature 464, 873–876 (2010)

    2010

  37. [37]

    Hu, M. et al. Memristor crossbar based hardware realization of BSB recall function. in Proc. Int. Joint Conf. Neural Networks. 1–7 (2012)

    2012

  38. [38]

    Park, S. et al. RRAM-based synapse for neuromorphic system with pattern recognition function. in IEEE Int. Electron Devices Meet. (IEDM) 231–234 (2012)

    2012

  39. [39]

    Alibart, F. et al. Pattern classification by memristive crossbar circuits using ex situ and in situ training. Nat. Commun. 4, 2072 (2013)

    2072

  40. [40]

    Mannocci, P. et al. In-memory computing with emerging memory devices: status and outlook. APL Mach. Learn. 1, 010902 (2023)

    2023

  41. [41]

    Guo, N. et al. Energy-efficient hybrid analog/digital approximate computation in continuous time. IEEE J. Solid-State Circuits 51, 1514–1524 (2016)

    2016

  42. [42]

    Govoreanu, B. et al. 10×10nm² Hf/HfOx crossbar resistive RAM with excellent performance reliability and low-energy operation. in IEDM Tech. Dig. 729–732 (2011)

    2011

  43. [43]

    Cai, Y . et al. High-density and high-reliability RRAM for memory and computing applications. in IEEE Int. Conf. Solid-State Integr. Circuit Technol. (2024)

    2024

  44. [44]

    Cowan, G. E. R. et al. A VLSI analog computer/math co-processor for a digital computer. in Proc. IEEE Int. Solid-State Circuits Conf. 82–83 (2005)

    2005

  45. [45]

    Array-based analog computation

    Kramer, A. Array-based analog computation. IEEE Micro 16, 40–49 (1996)

    1996

  46. [46]

    Chakrabartty, S. et al. Sub-microwatt analog VLSI trainable pattern classifier. IEEE J. Solid-State Circuits 42, 1169–1179 (2007)

    2007

  47. [47]

    Ielmini, D. et al. In-memory computing with resistive switching devices. Nat. Electron. 1, 333–343 (2018)

    2018

  48. [48]

    Punzenberger, M. et al. A new 1.2 V BiCMOS log-domain integrator for companding current-mode filters. in Proc. IEEE Int. Symp. Circuits Syst. 125–128 (1996)

    1996

  49. [49]

    Cong, Y . et al. Performance association and discrimination between current-mode and voltage-mode operation of monolithic linear circuits. in Proc. Southwest Symp. Mixed- Signal Des. 131 (2001)

    2001

  50. [50]

    Sun, Z. et al. Time complexity of in-memory solution of linear systems. IEEE Trans. Electron Devices 67, 2945–2951 (2020)

    2020

  51. [51]

    Zoppo, G. et al. Analog solutions of discrete Markov chains via memristor crossbars. IEEE Trans. Circuits Syst. I 68, 4910–4923 (2021)

    2021

  52. [52]

    Mannocci, P. et al. A generalized block-matrix circuit for closed-loop analog in-memory computing. IEEE J. Explor. Solid-State Comput. Devices Circuits 9, 47–55 (2023)

    2023

  53. [53]

    Achour, S. et al. Noise-aware dynamical system compilation for analog devices with Legno. in Proc. Int. Conf. Archit. Support Program. Lang. Oper. Syst. 149–166 (March 2020)

    2020

  54. [54]

    Udayanga, N. et al. Continuous-time analog computing circuits for solving the electromagnetic wave equation. in Proc. IEEE Int. Symp. Circuits Syst. 1–5 (May 2018)

    2018

  55. [55]

    Udayanga, N. et al. Continuous-time algorithms for solving Maxwell’s equations using analog circuits. IEEE Trans. Circuits Syst. I 66, 3941–3954 (2019)

    2019

  56. [56]

    Udayanga, N. et al. A radio frequency analog computer for computational electromagnetics. IEEE J. Solid-State Circuits 56, 440–454 (2021)

    2021

  57. [57]

    Huang, Y . et al. Analog computing in a modern context: a linear algebra accelerator case study. IEEE Micro 37, 30–38 (2017)

    2017

  58. [58]

    Hasler, J. et al. Continuous-time configurable analog linear system solutions with transconductance amplifiers. IEEE Trans. Circuits Syst. I 68, 765–775 (2021)

    2021

  59. [59]

    Ulmann, B. et al. Solving systems of linear equations on analog computers. in Proc. Kleinheubach Conf. 1–4 (2019)

    2019

  60. [60]

    Wang, Z. et al. Resistive switching materials for information processing. Nat. Rev. Mater. 5, 173–195 (2020)

    2020

  61. [61]

    Lanza, M. et al. Memristive technologies for data storage, computation, encryption, and radio-frequency communication. Science 376, eabj9979 (2022)

    2022

  62. [62]

    Resistance switching memories are memristors

    Chua, L. Resistance switching memories are memristors. Appl. Phys. A 102, 765–783 (2011)

    2011

  63. [63]

    Yao, P. et al. Fully hardware-implemented memristor convolutional neural network. Nature 577, 641–646 (2020)

    2020

  64. [64]

    Wan, W. et al. A compute-in-memory chip based on resistive random-access memory. Nature 608, 504–512 (2022)

    2022

  65. [65]

    Li, W. et al. A 40-nm MLC-RRAM compute-in-memory macro with sparsity control on- chip write-verify and temperature-independent ADC references. IEEE J. Solid-State Circuits 57, 2868–2877 (2022)

    2022

  66. [66]

    Ansari, M. S. A single-layer asymmetric RNN with low hardware complexity for solving linear equations. Neurocomputing 489, 1–12 (2022)

    2022

  67. [67]

    Revisit the analog computer and gradient-based neural system for matrix inversion

    Zhang, Y . Revisit the analog computer and gradient-based neural system for matrix inversion. in Proc. IEEE Int. Symp. Mediterr. Conf. Control Autom. Intell. Control. 1411– 1416 (2005)

    2005

  68. [68]

    Sun, Z. et al. Fast solution of linear systems with analog resistive switching memory (RRAM). Proc. IEEE Int. Conf. Rebooting Comput. 1–5 (March 2019)

    2019

  69. [69]

    Donatella, K. et al. Thermodynamic natural gradient descent. arXiv preprint arXiv:2405.13817 (2024)

    work page arXiv 2024

  70. [70]

    Bartosik, P.-L. et al. Thermodynamic algorithms for quadratic programming. arXiv preprint arXiv:2411.14224 (2024)

    work page arXiv 2024

  71. [71]

    Sun, Z. et al. In-memory eigenvector computation in time O(1). Adv. Intell. Syst. 2, 2000064 (2020)

    2020

  72. [72]

    Sun, Z. et al. One-step regression and classification with cross-point resistive memory arrays. Sci. Adv. 6, eaay2378 (2020)

    2020

  73. [73]

    Wang, S. et al. Optimization schemes for in-memory linear regression circuit with memristor arrays. IEEE Trans. Circuits Syst. I 68, 4900–4909 (2021)

    2021

  74. [74]

    Mannocci, P. et al. A universal analog in-memory computing primitive for linear algebra using memristors. IEEE Trans. Circuits Syst. I 68, 4889–4899 (2021)

    2021

  75. [75]

    Mannocci, P. et al. An analogue in-memory ridge regression circuit with application to massive MIMO acceleration. IEEE J. Emerg. Sel. Topics Circuits Syst. 12, 952–962 (2022)

    2022

  76. [76]

    Mannocci, P. et al. In-memory principal component analysis by crosspoint array of resistive switching memory: a new hardware approach for energy-efficient data analysis in edge computing. IEEE Nanotechnol. Mag. 16, 4–13 (2022)

    2022

  77. [77]

    Mannocci, P. et al. In-memory principal component analysis by analogue closed-loop eigendecomposition. IEEE Trans. Circuits Syst. II 71, 1839–1843 (2024)

    2024

  78. [78]

    Hong, C. et al. Solving All Eigenpairs With Resistive Memory-Based Analog Matrix Computing Circuits. IEEE Trans. Circuits Syst. I 73, 2522–2532 (2026)

    2026

  79. [79]

    Luo, Y . et al. Modeling and mitigating the interconnect resistance issue in analog RRAM matrix computing circuits. IEEE Trans. Circuits Syst. I 69, 4367–4380 (2022)

    2022

  80. [80]

    Luo, Y . et al. Smaller, faster, lower-power analog RRAM matrix computing circuits without performance compromise. Sci. China Inf. Sci. 68, 122402 (2025)

    2025

Showing first 80 references.