---
title: "Modern reconfigurable analog computers target differential equations, not neural networks, and their claimed edge is a solution time independent of problem size"
type: "claim"
status: "seedling"
audit_status: "capture-verified"
writer_model: "claude-opus-4-8"
flags: ["[unverified-quant] The 'constant solution time regardless of problem size' advantage is stated only as an asymptotic shape; the cited paper gives no joules or seconds benchmarking an analog solver against a digital one. Kept seedling; verification routed to [[question-verify-reconfigurable-analog-computer-ode-benchmark]]."]
source_url: "https://arxiv.org/html/2510.25942"
source_title: "Reconfigurable Analog Computers"
source_author: "arXiv:2510.25942, 'Reconfigurable Analog Computers' (Anabrid/SpriND reconfigurable-analog lineage, Bernd Ulmann; individual authors not recorded in capture)"
source_date: "2025-10"
source_quote: "the solution time on an analog computer is basically constant, regardless of the number of computing elements required"
source_tier: 1
provenance: "Promotion from 10-inbox/raw/2026-07-11-hop-analog-nn-inference-silicon.md, 2026-07-11"
origin: "hop-batch"
derived_from: ["10-inbox/raw/2026-07-11-hop-analog-nn-inference-silicon.md"]
date_created: "2026-07-11T00:00:00.000Z"
tags: ["analog-computing","differential-equations","ai-hardware","reconfigurable-analog","in-memory-compute","cross-domain-bridge","hardware-history"]
related_notes: ["claim-lukyanov-water-integrator-1936-hydraulic-differential-equation-solver","claim-hydraulic-analog-computers-independently-invented-economics-engineering","claim-encharge-en100-switched-capacitor-in-memory-compute"]
---


The serious primary literature on the 2020s analog-computer revival — reconfigurable analog machines such as Anabrid's **LUCIDAC** and **REDAC** — is about **dynamical systems and differential equations**, not about neural networks. The arXiv paper *Reconfigurable Analog Computers* (2510.25942, Oct 2025) frames these devices as co-processors for coupled and nonlinear ordinary differential equations, and never mentions neural-network inference.

The measurable edge the paper claims is asymptotic rather than a headline efficiency number: "the solution time on an analog computer is basically constant, regardless of the number of computing elements required," whereas digital solution time "typically grows with problem size, often much worse than linearly." Because an analog computer integrates the governing equations physically and in parallel — every computing element settling simultaneously — adding more state variables does not lengthen the wall-clock solve the way it does for a sequential digital integrator. This is a claim about the *shape* of the scaling curve, not about joules or seconds: the paper gives no benchmarked energy or time comparison, so the "measurably worse" framing for digital remains `[unverified-quant]` (see [[question-verify-reconfigurable-analog-computer-ode-benchmark]]).

This continues a lineage the vault already tracks in its historical form. Lukyanov's [[claim-lukyanov-water-integrator-1936-hydraulic-differential-equation-solver|1936 Water Integrator]] solved partial differential equations by letting water levels stand for variables, and it sits in a [[claim-hydraulic-analog-computers-independently-invented-economics-engineering|convergent-invention pair]] with Phillips's MONIAC. The reconfigurable electronic analog computer is the same idea — compute by physical analogy to the equations — modernized into a programmable co-processor.

Crucially, this differential-equation lineage is **distinct** from the analog-*inference* lineage (see [[claim-encharge-en100-switched-capacitor-in-memory-compute]]). They are different companies, different mechanisms, and different problems; press coverage of an "analog revival" tends to conflate the two.

> [!note] Seek's commentary:
> The useful discipline here is refusing to let "analog is measurably better" smuggle in a number the source doesn't provide. The paper's honest claim is about the scaling *curve*, not a benchmark — and that distinction is exactly what the [unverified-quant] flag preserves. The two-lineages point (math-analog vs AI-analog) is the capture's most genuinely non-obvious observation, and it is what keeps this note from being read as evidence that analog is winning at neural inference. It isn't; that's a separate, thornier story.
> — Seek