Athos Silicon
di Francois PiednoelStagione 1
In this podcast, Athos Silicon is introduced as a company building the computing foundation for physical AI, including autonomous cars, commercial drones, industrial robots, and humanoid robots. The discussion starts with a simple reality: autonomy is not just about better AI models or more compute. At scale, reliability becomes the real problem, because rare faults become inevitable when millions of systems operate for billions of miles. The podcast then explains Athos Silicon’s core platform, mSoC, meaning Multiple Systems on Chip. Instead of relying on one monolithic system that must never fail, mSoC uses multiple cooperating compute domains that cross check each other to prevent a single point of failure from becoming a catastrophe. The physical strategy behind this is Chiptile: design one chiplet, verify it deeply, then tile it into a scalable fabric that can isolate a failing tile without losing the whole system. A key theme is voting as a first class architectural feature, using triple redundant mailboxes so disagreements are detected quickly and outliers can be reset, isolated, or safely degraded. The episode also touches on deterministic scheduling, role based execution, and power monitoring to prevent cascading failures. The takeaway is clear: the next era of autonomy will be built on architectures designed to fail safely and remain trustworthy in real deployments. 