Adapted AI Edge Solutions according to the application level.
From Ultra Low Power to medium and complex Applications
- Digital neural network based on the Chain of HW Neuron cells all interconnected
- AI processors for ML and Edge computing
- AI Embedded Supercaculator for Deep Neural Networks.
1 – NeuroMem neuromorphic AI chip
The Neuromorphic chip is a bank of identical neuromorphic memory cells (neurons) which react to digital stimuli and can learn & recognize in real-time.
Parallel architecture
The NeuroMem unique parallel architecture allows the design of low-power and high-speed AI chips with unlimited neural network capacity:
- Neuron cell = memory + processing logic
- Neuron cells are identical and interconnected
- Neurons are accessed simultaneously in broadcast mode
- Neurons work in parallel without the need for a supervisor
Interchip Cascadability
The chip is a chain of 576 neurons and it can be connected with other external chips constituting a daisy-chain, thus increasing the capacity in number of neurons (1152, 1728, and so on …), which depends on the application and variability of data.
For 576 neurons x 256 synapses per neuron = 147 456 synaptic connections of 8 bits.
Its Neurons can learn and recognize patterns extracted from any data sources, such as images, audio waveform, bio signals, text and more, with less energy and complexity than modern microprocessors.
