Making Analog Training Scale: Co-Designing Mapping, Optimizer, and Converters

arXiv:2609.36584v1 Announce Type: new Abstract: Analog in-memory computing (AIMC) offers an alternative for model training by executing matrix operations directly where weights are stored. However, scaling AIMC to train modern deep models remains an open challenge due to severe hardware non-idealities, including physical weights with finite dynamic range and write granularity, analog-digital converters with finite resolution, and noisy and asymmetric updates. Guided by the insight that gradient accumulation is sensitive to precision and rounding errors, we adopt a mixed-precision training paradigm: executing forward and backward matrix multiplications in the analog domain while computing weight gradients in the digital domain. To enable scalable training, we present a holistic system-algorithm co-design that co-optimizes weight mapping to ensure well-conditioned physical and logical weight profiles, couples a preconditioned optimizer with threshold-triggered open-loop pulsing to stabilize training trajectories, and aligns converter dynamic ranges to suppress quantization errors. Evaluated via hardware-calibrated architectural simulations calibrated with electrochemical RAM measurements, our framework scales Transformer training up to 123ext{M} parameters with validation loss scaling as Lpropto N^{-0.231}, where N is the parameter count, comparable to Lpropto N^{-0.238} for digital training.

Related

• Physical Muon: Orthogonalization as an Equilibrium Computation
• Memory-Efficient LLM Pretraining via Minimalist Optimizer Design
• Activation Compression in LLMs: Theoretical Analysis and Efficient Algorithm
• MuonQ: Enhancing Low-Bit Muon Quantization via Directional Fidelity Optimization

Source: arXiv cs.LG | 2026-09-30