The conference program for June’s Design Automation Conference (DAC) in San Francisco has gone online, with machine learning forming a key part of the keynote tracks and the technical sessions.
Following a shorter Sky Talk given by Anna-Katrina Shedletsky, founder of Instrumental, on machine learning in manufacturing on Monday afternoon (June 25), IBM vice president of AI Dario Gil will take to the stage on Tuesday in a morning keynote to talk about the new types of computing device that will be needed for machine learning in the future. According to Gil, they include approximate computing and forms of quantum computing. He aims to highlight how these types of design will challenge the EDA community.
Following on from the IoT focus of previous years, Monday morning’s keynote given by Sarah Cooper, GM of IoT solutions at Amazon Web Services, will look at the use of embedded AI and connectivity in personal devices and whether further technological enhancements are needed to drive that market. Thursday’s keynote by Professor Mata Mataric from the University of Southern California at Los Angeles looks at another embedded application of AI: socially assistive robots.
On Wednesday, RISC-architecture luminary Professor David Patterson from the University of California at Berkeley, will describe the new generation of domain-specific accelerators that he sees being built on top of the open-source RISC-V instruction set architecture (ISA) – many of which are likely to go into machine-learning applications. Monday’s designer-track sessions include one that focuses on three industrial projects to build custom AI accelerators.
Before Patterson takes to the stage, PR ‘Chidi’ Chidambaram, vice president of engineering at Qualcomm, will take on another looming challenge – the rollout of 5G wireless services. The chips that go into these systems are expected to last longer than their predecessors as the replacement rate of telecom equipment falls, so he will call for the idea of design for durability – in an environment where nanometer-technology devices are more likely to suffer in-field failures from a variety of aging effects.
Monday’s tutorial sessions include two half-day sessions looking at different aspects of AI. One focuses on the use of machine learning in EDA flows; the other on deploying low-energy algorithms for use in embedded devices. Other conference sessions on machine learning look at dynamic reconfiguration, in-memory computing, and the verification challenges of deep learning.