- The framework groups differential-equation solving, matrix-equation solving and matrix-vector multiplication.
- Reported robotics demonstrations remain a company plan, rather than a completed deployment.
Defining a matrix-computing framework
Peking University researcher Sun Zhong is helping frame modern analog computing around matrix operations relevant to AI and communications, according to a November 3 QbitAI report. The article discusses a review involving researchers from Peking University, IBM's Zurich laboratory and Unconventional AI.
It describes the approach as mapping mathematical operations onto physical circuits, with resistive-memory devices providing programmable conductance. The review groups three core operations: solving differential equations, solving matrix equations and matrix-vector multiplication.
Working with digital systems
QbitAI describes the architecture of the chip company associated with Sun as using dedicated analog cores for matrix operations, with converters returning results to digital form. Communication between cores and chips, together with nonlinear functions, remains digital. The account presents this as cooperation with existing digital infrastructure.
The company expects to demonstrate applications in embodied-AI motion control and planning within a year, the report says. That remains a forward-looking plan. This report has not independently tested the hardware, and the framework should not be read as evidence of a commercially deployed replacement for GPUs or a general performance advantage across AI workloads.
Sources & context
Go to the original material. Company claims remain attributed to their sources.
01
