Abacus Vi SS2801
A deep-learning accelerating IC could be mainly applied to accelerating DCNN in machine vision.
-
It could finish the operations of
convolution, mean pooling and maximum pooling in CNN. And it supports MLP (multilayer
perceptron model), especially to optimize
small convolution filters, more suitable for modern machine vision algorithms.
By
optimizing data reuse rate, it could satisfy the request of operation with
little data bandwidth.
Its clock speed is 400MHz, and with 200GOPS of performance.
Working power
consumption (typ) 100mW.
Application
Scenarios: machine vision.
