Abstract

Automobiles are undergoing a dramatic transformation. Design efforts once focused mechanical powertrains and body styling are shifting to include high-speed digital processing supporting a diverse set of sensors and actuators. As autonomous cars move further toward autonomy, the processing becomes increasingly like that of advanced avionics and defense systems. This has created the need to re-think traditional approaches to SoC design and architecture.

We introduce an intelligent interconnect tools and architecture framework targeted for solving the needs of next-generation autonomous SoCs. Machine learning provides a solution for optimizing interconnect, where algorithms explore patterns in data and iterate solutions quickly. A requirements-driven design flow specifies IP blocks, their basic connectivity, performance requirements, and system-centric use cases. Instead of architecting the interconnect up front using a best-guess approach, NetSpeed’s intelligent interconnect uses machine learning to find optimum solutions. Each use case is evaluated for performance, power, area (PPA), FuSa requirements, and a NoC implementation is suggested. Architects can take a look at those automatically generated results and fine-tune their configuration accordingly. Along with PPA metrics, the FIT rate and achievable ASIL levels for ISO-26262 standards are also presented as output to the architects.

Biography

Rocco Jonack holds a master in EE from the RWTH Aachen and has more than 20 years of experience in SoC design, ranging from HW design, and architectural modeling to virtual platform modelling. He recently joined NetSpeed. Prior to that Rocco worked at Intel on modeling of VP models for SW validation. Before Intel he has been working with Sonics on interconnect modeling and implementation. In his earlier career he worked for Synopsys and Sican in various positions.