Digital Accelerators
Due to its sequential data movement between processor and memory unit, the traditional Von Neumann architecture leads to various limitations in performance.
Traditional analog in-memory computing
Analog in-memory computing (AIMC) technology addresses challenges in traditional digital accelerators such as memory bottleneck and energy inefficiency. However, it encounters other challenges such as retention and endurance.
AnalogAI Technology
Our innovative technology introduces novel synaptic devices using a new material with low leakage current and capacitor to solve the retention issue in capacitor based charge storage synapses.
New synaptic device and co-optimized algorithm to eliminate the limitations of traditional analog in-memory computing (AIMC)
Massively parallel inference and on-chip training
Linear and symmetric weight update for accuracy and stability
Low leakage current device and capacitor for longer retention