US 12,423,054 B2
Apparatus and methods for neural network operations supporting fixed point numbers of short bit length
Yunji Chen, Beijing (CN); Shaoli Liu, Beijing (CN); Qi Guo, Beijing (CN); and Tianshi Chen, Beijing (CN)
Assigned to CAMBRICON TECHNOLOGIES CORPORATION LIMITED, Beijing (CN)
Filed by Cambricon Technologies Corporation Limited, Beijing (CN)
Filed on Mar. 1, 2022, as Appl. No. 17/683,817.
Application 17/683,817 is a continuation in part of application No. 16/174,100, filed on Oct. 29, 2018, granted, now 11,295,196.
Application 16/174,100 is a continuation in part of application No. PCT/CN2016/081839, filed on May 12, 2016.
Claims priority of application No. 201610282480.4 (CN), filed on Apr. 29, 2016.
Prior Publication US 2022/0308831 A1, Sep. 29, 2022
This patent is subject to a terminal disclaimer.
Int. Cl. G06F 7/483 (2006.01)
CPC G06F 7/483 (2013.01) 26 Claims
OG exemplary drawing
 
1. An apparatus for neural network operations, comprising:
a fixed-point number converter configured to convert one or more first floating-point numbers to one or more first fixed-point numbers in accordance with at least one format;
a neural network processor configured to process the first fixed-point numbers to generate one or more process results, and
a floating-point number analyzing processor configured to determine the at least format of the fixed-point numbers by statistically analyzing one or more categories of the first floating-point numbers to determine a distribution pattern for each of the one or more categories over one or more data ranges.