Over the years, chips used in devices such as cars, medical equipment and cameras have become intelligent, (re)active and self-learning. Those features require a lot of computing power. In Eindhoven, GML researches and develops chips that have been inspired by the human brain. Our brain becomes smarter the more it learns, like a baby’s brain until the age of about 25 years. In addition, GML’s AI technology is so-called ‘on edge’, which runs in a device rather than in the cloud.
“The brain is the most efficient computer,” says Menno Lindwer, GML’s Vice President of IP and Silicon. The company’s name, GrAI Matter Labs, is a reference to the grey substance of the brain. “With our brain we can do things that no other computer in the world can.” Lindwer, who heads R&D in Eindhoven, calls GML’s approach ‘a break-through technology’. Two years ago, High Tech Campus in Eindhoven, where GrAI Matter Labs is based, labeled them ‘one of the most exciting’ startups of the Campus.
The brain consists of neurons and synapses. Neurons actually are tiny calculation units and our brain has about 80 billion of them, with each of them having hundreds of connections to other neurons. At present, it is not yet possible to put 80 billion similarly connected units on a chip. In terms of power the brain consumes about 20 Watt, less than a laptop processor, Lindwer explains. That is why GML is creating neuromorphic, brainlike, processors.
GML’s critical innovations need to be protected, with many such inventions continuously being patented. These patents are mostly related to data and neuroporphic processing, but also to new neuron models and image classification. Such procedures typically take several years until patents are granted.
Brabant and Paris
In 2016, GrAI Matter Labs SAS was founded in Paris, based on research which was performed at a department of UPMC, now part the Sorbonne university. The department was researching the perception of the human eye and brain. Based on that knowledge, it considered making a ‘smart’ eye or a chip which does calculations similar to the human brain. GML is a French company but the Paris area does not have a chip development ecosystem like in Noord-Brabant. Lindwer believes Brainport Eindhoven is an excellent location for R&D and chip-related engineering.
Eindhoven has a very open tech ecosystem, which has all the components that one needs to build up a chip company. The infrastructure, the suppliers, the people with the knowledge and the talents. They are all here.Menno Lindwer, GML’s Vice President of IP and Silicon
The ties of GML’s employees with the Technical University of Eindhoven (TUe are also strong), because many of them come from the region, many studied there and all have worked in high tech. At least a third of the new recruits has graduated from TUe, explains Lindwer, who also studied at this university.
“People that graduate from TUe aren’t scientists with fluffy stories. They are innovative people, but with a real engineering background that know how to make a product”, Lindwer says. “They don’t just look at that one technical solution. Most people that graduate from TUe have a broader view.”
Also part of the ecosystem in and around Eindhoven is the BOM, which has helped GML in many aspects. The BOM has organized events to inform GML about the ecosystem. The contacts from these meetings have eventually led to new customers for the company. Also, the BOM has made it easier for GML to get work permits for some of its new hires and it has facilitated access to cash. In past years, GML has secured funding of over €25 million by itself.
There are all kinds of companies that create smart AI technology inspired by the brain. It typically takes four years to develop a chip and none of these companies has yet finalized a product, says Lindwer. GML’s biggest competitors are not the companies with the radical new ideas, but rather the incumbents who currently still market AI processing devices based on generic high-power architectures.
Many users want to work with different AI processing companies, because devices from the incumbents are too slow and use too much power. GML’s technology is 10 to 100 times more efficient, claims Lindwer. The edge technology ensures that the chips in devices do not require external cloud computing. Therefore, future smart devices will not need to spend time and power on exchanging large amounts of data, needed for AI computations, with remote cloud servers. Not only will this improve power consumption and responsiveness of edge devices themselves, it will also strongly reduce power consumption in the big server parks.
If new customers in Europe are interested, they will likely talk to GML’s people in Paris, where the one of GML’s Business Development departments is based. GML also has staff in Silicon Valley (San Jose) because that is the place to be for any chip developer to be taken seriously, says Lindwer. However, the majority of GML’s staff works in Eindhoven and works in R&D.
With its technology, GML could serve any industry. “We can be of use for any sector that is now or at some point in the future dependent on Artificial Intelligence. Especially if AI runs in the device itself rather than in the cloud.” Because of its relatively small size, GML has to choose on which sector it will focus its business development. After long considerations, it has decided to focus on Robotics and Industrial Automation, industries that are most open to its solutions and that are highly developed in Noord-Brabant.
That does not imply that GML will not be open to getting customers from different industries. Life science & Health, of which Noord-Brabant is the European hotspot, is a sector that Lindwer could imagine working with, for instance in Medical Ultrasound, or with Surveillance or Automotive.