US 11,734,214 B2
Semi-programmable and reconfigurable co-accelerator for a deep neural network with normalization or non-linearity
Stephen Sangho Youn, Bellevue, WA (US); Steven Karl Reinhardt, Vancouver, WA (US); Jeremy Halden Fowers, Seattle, WA (US); Lok Chand Koppaka, Bellevue, WA (US); and Kalin Ovtcharov, Snoqualmie, WA (US)
Assigned to MICROSOFT TECHNOLOGY LICENSING, LLC, Redmond, WA (US)
Filed by Microsoft Technology Licensing, LLC, Redmond, WA (US)
Filed on Mar. 25, 2021, as Appl. No. 17/212,751.
Claims priority of provisional application 63/144,090, filed on Feb. 1, 2021.
Prior Publication US 2022/0245083 A1, Aug. 4, 2022
Int. Cl. G06F 13/40 (2006.01); G06N 3/04 (2023.01)
CPC G06F 13/4027 (2013.01) [G06N 3/04 (2013.01)] 19 Claims
OG exemplary drawing
 
1. A configurable stacked architecture for a fixed function datapath for use with an accelerator to accelerate an operation of a deep neural network (DNN), comprising:
a plurality of configurable micro-scalar processing units (SPUs) that perform at least one scalar operation on vector values from a received vector; and
a plurality of configurable micro-multi-functional units (MFUs) that perform vector operations on the vector values, wherein the plurality of configurable micro-SPUs and the plurality of configurable micro-MFUs are placed in an order to perform the operation of the DNN, wherein the order includes:
an output of a first micro-SPU of the plurality of configurable micro-SPUs being provided as an input to a first micro-MFU of the plurality of configurable micro-MFUs, and
an output of the first micro-MFU being provided as input to a second micro-SPU of the plurality of configurable micro-SPUs.