Title:
IN-MEMORY SPIKING NEURAL NETWORK BASED ON CURRENT INTEGRATION
Document Type and Number:
WIPO Patent Application WO/2022/057222
Kind Code:
A1
Abstract:
An in-memory spiking neural network (SNN) based on current integration. Calculation based on a charge domain is naturally compatible with the working mechanism of neurons. On the one hand, in order to avoid the non-ideality of using an NVM material, a memory cell of a synapse array in the architecture is a silicon-based SRAM cell. In addition, the provision of a modified NVM cell can also benefit from the architecture of the in-memory SNN. When the synapse array uses an SRAM cell as a storage cell, the design of a post-neuronal circuit corresponds to the same, so that the in-memory SNN architecture can be used for calculation of a multi-bit synaptic weight, and the number of columns of a combination is programmable. Further, in order to improve the area usage efficiency and save energy efficiency, in calculation of the multi-bit synaptic weight, the circuit is designed as a time multiplexing form of resource sharing. Finally, an automatic calibration circuit is provided to counteract changes in the on-current caused by factors such as process, voltage, temperature (PVT), so that the calculation result is more accurate.
Inventors:
YANG MINHAO (CN)
LIU HONGJIE (CN)
LIU HONGJIE (CN)
Application Number:
PCT/CN2021/081340
Publication Date:
March 24, 2022
Filing Date:
March 17, 2021
Export Citation:
Assignee:
REEXEN TECH CO LTD (CN)
International Classes:
G06N3/063
Foreign References:
| CN110543933A | 2019-12-06 | |||
| US20190370640A1 | 2019-12-05 | |||
| CN109165730A | 2019-01-08 | |||
| CN110852429A | 2020-02-28 | |||
| US20180260696A1 | 2018-09-13 | |||
| CN202010965425A | 2020-09-15 | |||
| CN111010162A | 2020-04-14 | |||
| CN109165730A | 2019-01-08 | |||
| CN103189880A | 2013-07-03 |
Other References:
See also references of EP 4024289A4
Attorney, Agent or Firm:
BEIJING GAOWO LAW FIRM (CN)
Download PDF: