Compressed Bent-Pyramid Format for Digital Stochastic Computing
A new digital in-memory stochastic computing architecture, DISCA, uses a compressed Bent-Pyramid format to address memory wall and Moore's Law limitations for AI at the edge.
Key facts
- DISCA is a digital in-memory stochastic computing architecture.
- It uses a compressed Bent-Pyramid format.
- The architecture targets AI applications at the edge.
- It addresses the memory wall and end of Moore's Law.
- Massive-scale AI models have millions or billions of parameters.
- AI applications are migrating to edge devices like robotics and UAVs.
- Analog and digital in-memory computing suffer from degradation.
- The paper is from arXiv:2511.17265.
Entities
—