In-memory computing circuit chip based on magnetic cache and computing device
By adopting an in-memory computing circuit based on magnetic cache in the deep learning in-memory computing architecture and using a timer to set the data retention time, the problems of limited SRAM capacity and high MRAM write power consumption are solved, and efficient in-memory computing is achieved. Data storage and energy efficiency improvements.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING HOUMO TECH CO LTD
- Filing Date
- 2021-07-16
- Publication Date
- 2021-10-22
AI Technical Summary
In the existing deep learning in-memory computing architecture, SRAM-based cache capacity is limited and static power consumption is high, which affects energy efficiency; while MRAM-based storage weight data has long write delays and high write power consumption, making it difficult to meet the in-memory requirements. Computationally efficient requirements.
An in-memory computing circuit based on magnetic cache is used, and the data retention time of the magnetic cache unit is set through a timer. The high density and capacity of the magnetic cache unit are used to cache the data output by the in-memory computing unit, and based on the data volume and Throughput rate adjusts hold time to reduce write latency and power consumption.
It effectively improves the data storage capacity and energy efficiency of in-memory computing, overcomes the high power consumption defect of long write delays in magnetic cache units, and achieves high-capacity cache with lower power consumption, which is suitable for different in-memory computing scenarios.