Feed Your Mind - HP Energy Efficient CiM and AI accelerators for sub-18Å Technologies - R. Krishnamurthy

#sscs #cas #ai #cim #hpc

This presentation will highlight some of the emerging challenges and opportunities for sub-18Å process machine learning and AI technologies in the rapidly evolving IoT industry. With Moore’s law process technology scaling well into the nano-scale regime, future SoC platforms ranging from high performance cloud servers to ultra-low-power edge devices will demand advanced AI capabilities and energy-efficient deep neural networks. New and emerging IoT markets for autonomous vehicles, drones, and wearables require even higher
performance at much lower cost while reducing energy consumption. Some of the prominent barriers to designing high performance and energy-efficient AI processors and SoCs in the sub-18Å technology nodes will be outlined. New paradigm shifts necessary for integrating special-purpose machine learning accelerators into next-generation SoCs will be explored. Emerging trends in SoC circuit design for machine learning and deep neural networks, specialized accelerators for digital and analog in-memory and near-memory computing, reconfigurable multi-precision matrix multipliers, ultra-low-voltage logic, memory and clocking circuits, AI inference accelerators including binary neural networks
and associated on-chip interconnect fabric circuits are described. Future braininspired neuromorphic computing circuit design challenges and technologies will also be reviewed. Specific chip design examples and case studies supported by silicon measurements and trade-offs will be discussed.