Yet beneath this boom lies an awkward foundational question, one that a new editorial in Frontiers in Neuroscience by Shantanu Chakrabartty of Washington University in St. Louis, Chetan S. Thakur of the Indian Institute of Science, and Andreas G. Andreou of Johns Hopkins University puts squarely at the center of the discipline: what, exactly, is neuromorphic about neuromorphic computing? The question sounds philosophical, but the authors argue it is anything but academic. Without a crisp statement of principle, they contend, the field risks dissolving into a loose family resemblance of brain-inspired buzzwords rather than maturing into a genuine computing paradigm.

The contrast with neighboring fields is stark. Classical computing rests on a precise abstraction—Boolean logic, the Turing machine, the von Neumann architecture—that is entirely indifferent to whether it is realized in relays, vacuum tubes, or modern fabrication technologies. Quantum computing likewise has a sharp essence: state vectors in Hilbert space, unitary evolution, entanglement, and measurement, a definition rigorous enough to support formal claims of quantum advantage. Neuromorphic computing has no comparable statement of principle. As the authors note, recent editorials have begun to sense the problem, most pointedly Giacomo Indiveri’s piece titled “Neuromorphic is dead. Long live neuromorphic,” which argues that the label has drifted far from its founding meaning.

History clarifies what the early pioneers actually intended. Carver Mead coined “neuromorphic” electronic systems in the late 1980s to describe circuits that exploit the physics of transistors operated in the subthreshold regime, where channel current depends exponentially on gate voltage in close analogy to ionic conductances in neuronal membranes. His 1990 manifesto argued that biology computes with ten-thousand-fold to billion-fold less energy than digital logic precisely because it uses device physics rather than fighting it. Around the same time, John Hopfield showed that computation could be a collective, emergent property of a dynamical system relaxing in an energy landscape, while David Hopfield’s collaborators Sejnowski and Hinton introduced statistical mechanics into neural computation—contributions now recognized with the 2024 Nobel Prize in Physics. Richard Feynman, who taught the famous course on the physics of computation alongside Mead and Hopfield, asked what physics itself can compute and where the thermodynamic limits of computation lie. What these pioneers intended was not the imitation of neurons but the grounding of computation in physical dynamics and energy.

From this vantage point, the authors identify a core misconception in today’s neuromorphic discourse. Statements such as “uses subthreshold CMOS,” “uses memristors,” or “uses neural processing units” are statements about implementation—about form, not essence. Just as “uses transistors” never defined classical computing, and “uses superconducting junctions” does not define quantum computing, a particular device technology cannot define neuromorphic computing. Following an idea introduced by the late David Marr, the authors argue that a computing principle must abstract away from the physical substrate on which it is implemented. If neuromorphic computing is to be a paradigm rather than a product category, its defining principle must survive translation across substrates: silicon, oxides, photonics, or even wetware.

So what is the substrate-independent essence? The authors’ most defensible candidate is the word “thermodynamic.” Nervous systems evolved under severe metabolic budgets: the human brain runs on roughly 20 watts, and cellular-level energy accounting shows that spiking, signaling, and synaptic transmission are all tuned to minimize metabolic cost. Insect brains sharpen the point dramatically. A honeybee forages over kilometers, navigates by landmark and sky compass, communicates, and learns with about a million neurons, while a fruit fly flies and adapts with roughly a tenth of that, on power budgets in the microwatt-to-milliwatt range. Direct measurements in fly photoreceptors show that the energy-information trade-off is enforced down to the level of single cells. The insect brain, the authors argue, is the gold standard for edge AI—full sensing, autonomy, and lifelong learning within a milliwatt budget—a benchmark that no current engineered system, neuromorphic or otherwise, comes close to matching.

In this framing, neuromorphic computation is, at its core, computing with and against thermodynamics: representing problems as energy functions and entropy production, letting physical dynamics perform descent, relaxation, and sampling, and treating statistical mechanics as the native mathematics of the machine. This dovetails with the emerging research program of thermodynamic computing, which takes fluctuations, dissipation, and stochastic dynamics as computational primitives rather than nuisances. It also connects to stochastic thermodynamics, where fluctuation theorems, uncertainty relations, and thermodynamic speed limits now make it possible to quantify the energy costs of computation in systems far from equilibrium, from digital circuits to brains. Crucially, the authors note that any constraint—structural, architectural, or physical—produces entropy or dissipates energy, so the essence of neuromorphic computing may be thermodynamic mechanisms that optimize parallel processing under physical constraints of energy, delay, and variability.

A thermodynamics-based framework, the authors argue, would first need to explain the four principles that many researchers currently treat as the fundamental signature of neuromorphic computing. The first is event-based, asynchronous representation, where computation, sensing, encoding, and communication are unified and information is carried by the timing of sparse events rather than by a clocked global state. The second is distributed, massive parallelism with memory and processing collocated within millions to billions of simple interconnected units. The third is analog processing, which extracts precision per joule from continuous representations that exploit device physics and locality. The fourth, and in the authors’ view most important, is noise-induced transitions and fluctuations, in which thermal and device noise—and possibly quantum fluctuations—drive escape from local minima, probabilistic inference, and exploration, making noise a computational resource rather than an error to be suppressed.

Architecturally, such a framework should provide a scientific rationale for machines built on multi-time-scale dynamics: fast membrane-like state variables, slower synaptic traces, and still slower homeostatic and structural processes, all coupled through local, continual learning rules and network oscillations. In such a machine there is no distinction between inference and learning—both are the same physical dynamics viewed at different time scales, with no separate training phase and no offline weight download. Everything observed at the system level, from polychronicity to predictive and efficient coding, becomes an emergent phenomenon. Synchronization across modular units, for instance, is not programmed but arises from coupling and possibly from constraints on entropy production, in the way oscillatory assemblies emerge across columns of cortex. Attention, likewise, is not a scheduled control signal but an emergent allocation policy, concentrating the metabolic budget where information matters most.

The authors’ final demand is the most sobering: the field should impose upon itself the same discipline that quantum computing did, articulating a clearly defined neuromorphic advantage measured against GPUs. They draw a sharp distinction between neuromorphic emulation and neuromorphic realization. Emulation reproduces neural dynamics—spikes, plasticity, asynchrony—on top of standard synchronous digital logic; realization lets the physics of the substrate itself carry the computation. Because scaling analog dynamics is hard and the sources of neuromorphic advantage remain poorly understood, most current platforms, including the largest spiking systems, are emulators built from digital circuits. Emulation is genuinely useful for algorithm exploration, neuroscience simulation, and low-power edge inference, but an emulator pays the digital tax on every spike, and it is far from clear that simulating stochastic analog dynamics with clocked logic can ever decisively outperform GPUs that are themselves superb digital simulators. If the field settles for emulation, the authors warn, it risks becoming an accelerator niche. If it pursues realization—computing with energy, dynamics, and noise as the brain does—it may yet become what Mead, Hopfield, and Feynman glimpsed: a third computing paradigm, with thermodynamics as its core logic.

Subject of Research: The foundational definition and thermodynamic principles of neuromorphic computing

Article Title: What is neuromorphic about neuromorphic computing?

Article References: Chakrabartty, S., Thakur, C. S., & Andreou, A. G. (2026). What is neuromorphic about neuromorphic computing?. Frontiers in Neuroscience, 20, Article 1980751. https://doi.org/10.3389/fnins.2026.1980751

Image Credits: AI Generated

DOI: 10.3389/fnins.2026.1980751

Keywords: neuromorphic computing, thermodynamics, spiking neural networks, Carver Mead, stochastic thermodynamics, energy efficiency, edge AI, memristors, Intel Loihi, IBM TrueNorth, quantum computing, unconventional computing