IBM Takes AI Chip Research to Next Level - RTInsights

IBM is developing a processor to improve system efficiency by combining compute and memory in a single device overcoming what is known as the Von Neumann bottleneck. IBM this week at an IEEE CAS/EDS AI Compute Symposium advanced an effort to improve the efficiency of systems by a factor of a thousand by 2029. It aims to accomplish this by giving developers access to an open-source Analog Hardware Acceleration Kit written in Python. The kit enables them to begin testing an approach to in-memory computing that will run neural networking algorithms much faster than any existing processor. The processor IBM is developing achieves that goal by combining compute and memory in a single device to overcome what is known as the Von Neumann bottleneck.