Abstract Self-timed circuits can significantly improve the energy efficiency of large-scale digital computer systems. The benefits of these circuits include switching activity only when there is useful work being done, and the ability to optimize for expected delays rather than the system-wide worst-case delay. These benefits are partially offset by extra signals introduced to implement synchronization and communication protocols. We describe both analytical results as well as design examples that demonstrate the efficiency of self-timed systems. The benefits of self-timed logic are immediately evident when designing large-scale neuromorphic systems—a potential approach towards intelligent machines. We highlight the uses of self-timed logic in state-of-the-art neuromorphic systems, and advocate for a quantitative, full-stack approach to evaluating the trade-offs in neuromorphic system design enabled by recent advances in design automation. This article is part of the discussion meeting issue ‘Bits, neurons and qubits for sustainable AI’.
Rajir Manohar (Thu,) studied this question.