From Synapses to Silicon: Theoretical Foundations and Neuromorphic Implementations of Biological Learning

Learning in biological neural systems emerges from the coordinated interaction of molecular, cellular, and circuit-level mechanisms that collectively give rise to adaptive behavior across timescales spanning milliseconds to years. At the most elementary level, synaptic efficacy is dynamically regulated through activity-dependent processes, including voltage- and calcium-mediated long-term potentiation (LTP) and long-term depression (LTD), spike-timing-dependent plasticity (STDP), and metaplasticity. These local synaptic rules, however, do not operate in isolation: neuromodulatory systems impose global signals that gate plasticity, encode reward prediction errors, and regulate the transition between exploratory and consolidative learning states. Understanding how local synaptic rules and global neuromodulatory signals are orchestrated across neural circuits to solve credit assignment, the fundamental problem of attributing outcomes to distributed neural causes, remains one of the central open questions in theoretical and systems neuroscience.Building upon these biological foundations, a growing body of theoretical work has sought to derive learning algorithms that are simultaneously computationally powerful and mechanistically plausible. Frameworks such as predictive coding, contrastive Hebbian learning, target propagation, and equilibrium propagation offer alternatives to error backpropagation that respect key biological constraints, including locality of synaptic updates, the absence of weight symmetry, and the use of sparse, asynchronous spike-based communication. Translating these theoretical insights into engineered systems is the central ambition of neuromorphic engineering.Analog VLSI implementations of conductance-based synaptic dynamics, subthreshold circuits emulating calcium-dependent plasticity, and mixed-signal neuromorphic processors with co-located memory and computation have demonstrated that key aspects of biological learning can be realized in silicon with orders-of-magnitude improvements in energy efficiency relative to conventional digital hardware. More recently, emerging post-CMOS substrates, including resistive switching (memristive) devices, phase-change memory, ferroelectric tunnel junctions, and electrochemical synapses, offer material-level analogs of synaptic weight storage and update dynamics, opening new avenues for dense, scalable, and inherently adaptive hardware. The integration of these device-level innovations into full on-chip learning architectures, supporting spike-based, event-driven computation with online weight updates and minimal off-chip communication, represents a frontier that demands close co-design between neuroscience theory, algorithm development, and hardware engineering.This Research Topic solicits original research articles, reviews, theory and methods papers that advance our mechanistic and implementational understanding of learning in neural systems. Contributions are invited across the following themes:- Theoretical frameworks for neural learning- Biologically plausible learning algorithms- Biophysical and computational models of synaptic plasticity- Neuromodulatory gating and three-factor learning rules- Analog and mixed-signal circuit implementations- Emerging non-volatile device technologies for synaptic emulation- On-chip learning architectures- Continual and online learning under hardware constraints- Benchmarking methodologies and evaluation frameworksContributions that establish explicit, quantitative mappings between biological learning mechanisms and their engineered analogs, whether through theoretical derivation, large-scale simulation, or physical implementation, are especially encouraged. Purely biological, molecular, or device-physics studies without a clear neuromorphic implementation or neural-inspired computing component fall outside this Topic’s scope. This Research Topic aims to serve as a definitive reference for the neuromorphic engineering community by tracing a coherent intellectual path from the molecular substrates of synaptic plasticity through to scalable, benchmarked, on-chip learning systems.Topic Editor Rajkumar Kubendran is a co-founder of TempoSense Inc. and holds shares and patents in the company. All other Topic Editors declare no competing interests with regard to the Research Topic subject. Topic editors: Prafull Purohit, Rajkumar Kubendran Source: https://www.frontiersin.org/research-topics/83913/from-synapses-to-silicon-theoretical-foundations-and-neuromorphic-implementations-of-biological-learning