Imec and KU Leuven present energy-efficient AI chip for robots
Researchers at imec and KU Leuven have designed the first ever chip that combines a digital and analog coprocessor to speed up computations for artificial intelligence. The DIgital and ANalog Accelerator, or DIANA for short, performs different types of calculations automatically in the most energy-efficient way. Among other things, DIANA can be used to allow robots to efficiently calculate which object is in their field of view and how to grasp it.
Artificial intelligence is rapidly entering our daily lives, from smartphones that recognize the face of their owners to robot vacuum cleaners that can avoid socks lying around. To enable meaningful applications that truly improve our lives, these devices must be able to interact with their environment even faster and better. The processing of environmental images, sounds and other data is best done on the device itself for privacy reasons, but it also requires a lot of computing power which can quickly drain the battery. Today, some vehicles and smartphones are already equipped with a digital processor specifically designed to speed up calculations for such artificial intelligence. However, to enable high-performance augmented reality glasses, an autonomous drone or a smart robot, a new generation of AI chips is needed that knows how to perform these calculations in a much more energy-efficient way.
Two years ago, researchers at imec developed a new chip architecture in which calculations are performed directly in computer memory through analog technology. This analog accelerator makes it possible to perform most of the operations ten to one hundred times more energy-efficient than in a digital accelerator. For another part of the operations, the calculation precision and programmability of a digital accelerator are better suited. To combine the best of both worlds, researchers from imec and KU Leuven, with support from the Flemish government through the Flemish AI Research Program, developed a processor that combines the analog and a digital coprocessor. This led to DIANA, a new AI chip manufactured by chip manufacturer GlobalFoundries.
The hybrid chip automatically performs the operations on the coprocessor best suited for each specific task. DIANA thus combines the advantages of an analog coprocessor (processing speed and energy efficiency) with the broad applicability of a digital coprocessor. This opens up a wide range of new applications that will be further investigated within the Flemish AI research program. For example, a robotic arm is being developed that can automatically recognize and grasp objects. The various calculation operations required for image recognition and robot control can thereby be carried out energy-efficiently on one and the same chip.
"The rapidly increasing heterogeneity in artificial intelligence algorithms requires new hardware that can perform the different types of computations performantly and energy efficiently. For some applications, such as pattern recognition, an analog coprocessor is best suited. Other applications, such as reasoning about those observations, additionally require a digital coprocessor. The solution we now propose combines the best of both worlds on one compact chip and always performs calculations in the most energy-efficient way, independent of the specific application," says Marian Verhelst, research director at imec and professor of microelectronics at KU Leuven.
How can we help?
The Leuven MindGate team is at your disposal for any questions about the Leuven Innovation Region. Do you want to invest, work or study in the region? We can help you find your way.
We also facilitate collaboration and innovation between companies, knowledge institutes and government within the Leuven Innovation Region, and we are happy to guide any of these stakeholders towards innovation.