Innatera, a spin-off of Delft University in the Netherlands, is developing a new breed of microprocessors that aim to bring brain-like intelligence to sensors. Adding intelligence at the edge is expected to enable a whole range of applications in the IoT, automotive, healthcare and other industries and potentially help usher in an age of privacy-preserving personalized AI on mobile devices.
The Dutch scale-up specializes in neuromorphic computing, a technology that aims to mimic how the human brain works, slashing the memory and power requirements of AI by orders of magnitude. The technology is gaining traction as the limitations and energy demands of traditional AI architectures become increasingly apparent. Rather than performing sequential operations on data stored in memory, neuromorphic chips use networks of artificial neurons that communicate through spikes, much like real neurons.
Innatera, another neuromorphic chip company called Gemesys, and Seth Bannon, a Silicon Valley investor at the venture capital firm Fifty Years, talked about this trend during a panel moderated by The Innovator’s Editor-in-Chief at the Hello Tomorrow conference in Paris on March 13.
Innatera CEO Sumeet Kumar envisions a future where neuromorphic chips increasingly handle AI workloads at the edge, while larger foundational models remain in the Cloud.
“The AI value chain is built on a lot of inefficiency,” he says. “Our aim is to make AI more accessible and more efficient.” Neuromorphic systems have distinct advantages, particularly for edge computing applications in industrial IoT and consumer devices, because they allow devices to make sense of data at the source without the need for large, power-hungry processors, he says.
If sensors could not just collect, but interpret, data at the edge predictive maintenance in factories could become significantly better, says Kumar.
Neuromorphic computing could also help the automotive sector introduce a host of new services. For example, intelligent radar sensors could differentiate between objects and people in a vehicle’s cabin and identify if a baby or child is forgotten in the car and alert the driver. They could also enable new types of gesture and voice recognition services that would allow drivers to reliably issue commands without having to fiddle with screens or the car’s controls to open a window or operate the entertainment system.
The combination of cameras and sensors with built in intelligence could additionally permit cars to hit the brakes faster than human drivers to avoid an accident. And if intelligence were to be distributed through sensors instead of centralized, vehicle systems would be able to respond faster, consume less energy, and operate more reliably in the event of failures, says Kumar.
Neuromorphic computing is expected to play a crucial role in enabling personalized AI services on wearables and mobile phones, says Kumar. With wearables like watches and Fitbit-like devices the key challenge is that much of the detection has to happen in the Cloud or when you explicitly place a finger on a sensor but aberrant heartbeats happen randomly,” says Kumar. “The future of devices is to be always on, continuously monitoring, a key to preventative care,” says Kumar.
Adding intelligence to sensors permits the monitoring to occur locally, it doesn’t need to be sent anywhere, preserving privacy, he says, increasing consumer willingness to apply AI to all areas of their lives.
Innatera is backed by the European Innovation Council as well as European deeptech VCs Matterwave Ventures, MIG Capital, InvestNL Deep Tech Fund, Delft Enterprises, and a group of private investors and entrepreneurs. It has raised 25 million euros so far.
Competitors include microcontroller vendors such as ST Microelectronics and Nordic Semiconductor, which make components that Innatera aim to displace, says Kumar.
Innatera also competes wit a growing number of companies developing AI accelerators, companion devices to microcontrollers that are already in use. The problem with these solutions is that they make the bill of materials for devices larger by adding an additional chip. “We have a different take,” says Kumar. “Our chip is the only one that is sitting next to the sensor, making systems smaller and much more efficient and effective,” he says.
Innatera says its differentiator with other neuromorphic computing startups, such as BrainChip, is that its approach offers efficiency improvements of 10x to 30x depending on the application, he says.
“Neuromorphic computing has immense potential,” says Kumar. “It will end up unlocking a new breed of applications that we’re probably not even thinking of today.”
This article is content that would normally only be available to subscribers. Sign up for a four-week free trial to see what you have been missing.
To access more of The Innovator’s Startup Of The Week articles click here.
