Fractal analog random access memory
By simulating random access memory devices using fractals, and utilizing sampling and holding capacitors in the array geometry, combined with readout transistors and peripheral amplifiers, the problem of slow matrix-vector multiplication speed was solved, enabling efficient training of artificial intelligence neural networks.
Patent Information
- Application Number
- CN202480078366.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-12-15
- Filing Date
- 2024-11-03
- Publication Date
- 2026-07-10
AI Technical Summary
In existing technologies for training artificial intelligence neural networks, matrix-vector multiplication operations are slow, resulting in low training efficiency and making it difficult to meet the requirements of high parallelism and low power consumption.
Using a fractal analog random access memory device, fast random access to memory cells is achieved through sampling capacitors and holding capacitors in the array geometry, combined with readout transistors, peripheral amplifiers and analog multiplexers, supporting highly parallel matrix-matrix multiplication operations.
It enables rapid and economical extraction of voltage values from large-scale capacitor arrays, improves the efficiency of matrix-matrix multiplication, and supports efficient training of artificial intelligence neural networks.
Smart Images

Figure CN122374824A_ABST