Hardware/software co-design attempts to optimize the hardware and software components of a complex electronic system while satisfying the project’s goals and design constraints, which usually encompass performance, cost, and power consumption.
This article discusses the five core principles that helped Recogni build a geographically diverse engineering organization that successfully developed a machine-learning (ML) chip while meeting ambitious performance, power-consumption, and development schedule goals. ML software is a rather extreme software case because of the non-deterministic nature of ML development. Experiences with this unique software niche underscore the critical nature of several of these five core principles.