Sean Kinzer

TL;DR: This paper defines Planaria1, a microarchitectural capability that can dynamically fission (break) into multiple smaller yet full-fledged DNN engines at runtime that enables spatially co-locating multiple DNN inference services on the same hardware, offering simultaneous multi-tenant DNN acceleration.

TL;DR: One possible 3D-stacked microarchitecture is defined, dubbed BiHiwe, that leverages clustering and hierarchical design to best utilize power-efficiency of the mixed-signal domain and 3D stacking and also builds models for noise, computational non-idealities, and variations.

TL;DR: BiHiwe as discussed by the authors proposes a wide bit-interleaved analog vector unit comprising of low-bitwidth multiply-accumulate modules that operate in the analog domain and share a single A/D converter.

TL;DR: PolyMath as discussed by the authors is a cross-domain stack that provides a unified computational stack across multiple, but not all, domains by defining a high-level crossdomain language (CDL), called PMLang, which encapsulates mathematical properties to be expressive across multiple domains.

TL;DR: A Bayesian optimization framework called Glimpse, incorporating the mathematical embedding of the hardware specification of the GPU accelerators dubbed Blueprint to better guide the search algorithm and focus on sub-spaces that have higher potential for yielding higher performance binaries.