Neuromorphic processor designer appoints EDA veteran as chair
Dutch neuromorphic processor designer Innatera has appointed Prof Alberto Sangiovanni-Vincentelli as Chairman of its Board of Directors.
Vincentelli is Innatera’s first Chairman of a board that includes deep-tech investors Christian Reitberger of btov Partners and Soren Hein of MIG Verwaltungs and Prof. Geert Leus of the Delft University of Technology.
The company has designed a sub-milliwatt signal processing chip using and event-driven spiking neural network architecture for edge applications. The company currently has 30 employees across its headquarters in the Netherlands and its new design centre in Bangalore, India and is a spin-off from the Delft University of Technology.
Neuromorphic, or spiking, neural networks, also called event-driven and optical flow AI, emulate the way the brain works and are a key areas of research and commercialisation of silicon in Europe for the low power capability. Prophesee in France has shipped the first camera to use the approach, while Opteran in the UK is also developing a low power imaging chip based on the technology.
Related neuromorphic processor articles
- Startup raises €5m for neuromorphic edge chip
- Optical flow AI startup Opteran raises £2.1m
- Prophesee in China partnership
- Open source library for event-driven AI vision
“In a market that is crowded by numerous semiconductor plays on AI, I am excited by Innatera’s differentiated approach to achieving ultra-low power intelligence at the sensor edge”, said Prof. Vincentelli. “Success for a chip company follows from a combination of disruptive technology, flawless execution and a great team, and I believe Innatera is in the sweet spot”.
Vincentelli helped to found the two leading Electronic Design Automation companies Cadence Design Systems and Synopsys and has served on the Board of Directors of Cadence, KPIT, Expert.ai and Cy4Gate, as well as the advisory boards of HP, STMicroelectronics, General Motors and United Technologies and as consultant to Mercedes, Intel and BMW.
Prof. Vincentelli is the Edgar L. and Harold H. Buttner Chair of Electrical Engineering and Computer Sciences at the University of California at Berkeley
“It is an honour for us to have Prof. Vincentelli on our board and a testament to the strong fundamentals Innatera is built on. He is no stranger to taking revolutionary new technologies to market, and brings a wealth of industry experience and wisdom,” said Sumeet Kumar, CEO of Innatera.
The chip uses a massively parallel array of spiking neurons and synapses to provide native acceleration for continuous-time spiking neural networks with fine-grained temporal dynamics. This allows the spiking neural net architecture to provide high-fidelity signal processing and pattern recognition in real time, within a sub-milliwatt power budget.
“Our characterization of the chip shows the excellent power-performance capabilities of analogue-mixed signal spiking compute technologies, and how much intelligence they will allow us to integrate into edge sensing applications,” said Amir Zjajo, Chief Scientist at Innatera.
The company has showed a tiny spiking neural network model performing real-time audio recognition with only 500 spike events per inference. Each spike consumes sub-picojoule energy, approaching the energy levels of biological neurons and synapses. Key to this, Innatera says, is its combination of the novel spiking approach and its ultra-low power analog-mixed signal silicon architecture.
“Innatera’s neuromorphic architecture operates at extremely low power, especially when compared with traditional deep-learning accelerators,” said Linley Gwennap, principal analyst at The Linley Group. “Unlike other analogue vendors, Innatera offers software to abstract the analogue fabric’s complexity, making it simple for the customer to use. This approach is well suited to bringing intelligence to low-power sensors.”
Innatera expects to sample to select customers in the second half of 2021. The company’s silicon engineering is complemented by the development of the dedicated software development kit needed for spiking networks and applications
Related neuromorphic articles
- First event-driven AI camera ships
- NASA looks to BrainChip’s spiking neural network chip for space
- Event driven sensor boosts drone safety
- European €3m project to boost spiking AI adds SynSense
- Eta adds spiking neural network support to MCU
- CEA-Leti integrates spiking neural network
Other articles on eeNews Europe
- Do patents kill innovation? The US patent office is asking.
- Nexperia spins out its semiconductor equipment business
- Flexible fab could slash chip shortage
- System-in-package consortium for heterogeneous chiplets
- Achronix cancels $2bn SPAC deal
- TSMC looks to 12nm Dresden fab
- European startup ships first commercial quantum processor
- Ascenium raises $16m for next generation software defined processor