Non-volatile memory-based activation function
Analog memory-based activation functions using non-volatile memory devices address inefficiencies in digital circuits by performing activation operations directly in the memory array, enhancing energy efficiency and computational speed in deep learning neural networks.
GB2638643BActive Publication Date: 2026-07-02INTERNATIONAL BUSINESS MACHINE CORPORATION
Patent Information
- Authority / Receiving Office
- GB · GB
- Patent Type
- Patents
- Current Assignee / Owner
- INTERNATIONAL BUSINESS MACHINE CORPORATION
- Filing Date
- 2023-11-07
- Publication Date
- 2026-07-02
AI Technical Summary
Technical Problem
Existing digital circuits for implementing activation functions in deep learning neural networks are inefficient in terms of energy consumption and require data movement, which hinders performance.
Method used
Implementing analog memory-based activation functions using non-volatile memory devices, such as memristive elements, to perform activation operations directly in the memory array, eliminating the need for data movement and reducing energy consumption.
Benefits of technology
This approach enhances energy efficiency and computational speed by performing activation functions in-place within the memory array, reducing the need for digital conversion and lowering overall energy usage.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure 00000001_0000 
Figure 00000002_0000 
Figure 00000003_0000
Abstract
Analog memory-based activation function for an artificial neural network can be provided. An apparatus can include at least two non-volatile memory devices connected in parallel such that the current
Need to check novelty before this filing date? Find Prior Art