The UIB develops a revolutionary AI chip with minimal power consumption

PalmA team of researchers from the Electronic Engineering Group (GEE) at the Balearic Islands Institute of Artificial Intelligence (IAIB) has achieved a significant breakthrough in the field of high-performance intelligent processing. The group has designed and manufactured a chip with an innovative artificial intelligence architecture, specifically designed for energy-critical environments such as IoT devices and edge computing systems. The chip uses stochastic computing to implement a morphological neural network, a technique that enables a high degree of parallelization within the circuit and significantly increases energy efficiency. According to the researchers, "the probabilistic coding used in the chip resembles the way neurons in our brain operate when communicating with each other, and this is key to achieving high energy efficiency," explains Josep Lluís Rosselló, principal investigator of the project.

Despite the limited precision of basic operations inherent in stochastic computing, when hundreds of thousands of these operations are chained together to achieve a final result—for example, recognizing an image or sound—the overall accuracy is virtually unaffected, as intermediate inaccuracies are self-correcting. This probabilistic nature allows the chip's supply voltage to be reduced below the usual margins for digital circuits.

At the forefront

Manufactured using 180nm CMOS technology and operating at 0.81 volts, the chip has been tested against the MNIST database of 70,000 handwritten digit images. The results are astonishing: it achieves an energy efficiency of over 1 TOPS/W with a consumption of less than 1 mW, surpassing many alternative architectures implemented with more advanced technologies.

"This advance opens the door to a new generation of smart devices with minimal power consumption and high computing capacity, essential for the development of the Internet of Things," highlights Miquel J. Roca Adrover, principal investigator of the GEE group. According to experts, this architecture is ideal for battery-powered devices, smart sensors, wearables and distributed monitoring systems. The project has been made possible thanks to the support of the Innova UIB-PYMES program, within the Espai Valida call, aimed at the maturation of R&D projects with proof of concept and the transfer of research results to the market.