Analog computing revival: resistive memory arrays solve equations faster
New paper reveals analog circuits can tackle matrix equations 10x more efficiently.
Researchers Sun et al. present a comprehensive survey on modern analog computing for solving differential and matrix equations. They identify three core primitives: solving differential equations, solving matrix equations, and performing matrix-vector multiplications. Among various hardware implementations, resistive memory arrays are noted as particularly promising due to their implementation efficiency. The paper surveys recent progress using advanced analog CMOS circuits and resistive memory arrays, and discusses applications for AI and scientific computing, along with precision, scalability, and links to in-memory computing.
- Identifies three core primitives: solving differential equations, matrix equations, and matrix-vector multiplications.
- Resistive memory arrays offer the most efficient analog implementation, enabling in-memory computation.
- Addresses precision and scalability issues with potential hybrid analog-digital solutions for real-world applications.
Why It Matters
Analog computing with resistive memory could slash energy use in AI training and scientific simulations by orders of magnitude.