A 'smart' chip that reduces both consumption and computing time marks a breakthrough in high-performance computing at Politecnico di Milano

Dramatically reducing energy consumption while accelerating the processing of large amounts of data. This is the aim of the new chip developed by a group of researchers from the Department of Electronics, Information and Bioengineering – DEIB at the Politecnico di Milano, led by Professor Daniele Ielmini, and presented in the study published in the prestigious journal Nature Electronics, with the researcher Piergiulio Mannocci as the first author.

The work originated as part of the ANIMATE (ANalogue In-Memory computing with Advanced device TEchnology) project, which was awarded an ERC Advanced Grant in 2022 and drew on Daniele Ielmini's preliminary research into CL-IMC (Closed-Loop In-Memory Computing) and the opportunities it could offer, i.e. faster solving of calculations and use of up to 5,000 times less energy than digital computers. The goal of the ANIMATE project was to develop the technology for the device, the circuits and system architectures, and the set of applications to validate the CL-IMC.

The resulting chip makes use of in-memory computing, which aims to overcome a limitation of computers: the need to continuously move data between the memory and the processor. By eliminating this internal 'traffic', the systems become faster and more energy efficient. In the recently published study, the DEIB team presented a fully integrated analogue accelerator for solving linear and non-linear systems of equations designed with CMOS (Complementary Metal-Oxide-Semiconductor) technology, implemented as standard for the production of silicon integrated circuits.

The device uses two 64×64 arrays of programmable resistive memories: an array is a kind of 'ordered grid' made up of identical parts arranged in rows and columns, similar to graph paper, where each intersection between a row and a column represents a memory cell. The cells are based on SRAM (Static Random-Access Memory) technology, a type of fast and stable memory which in this case is combined with integrated resistors for programming different levels of resistance. The architecture is complemented by an innovative model of analogue processing that utilises components integrated in the chip such as operational amplifiers and analogue-to-digital converters.

The set allows the system to handle complex calculations directly in the structure of the memory, avoiding the need to move data to an external processor, thereby reducing calculation times to a significant extent. In tests, the chip achieved similar accuracy to conventional digital systems, but with lower power consumption, less computing latency and a smaller footprint on the silicon.

Daniele Ielmini, a member of DEIB and head of the research group, explained: "The integrated chip demonstrates the feasibility on an industrial scale of a revolutionary concept such as analogue computation in memory. We are already working on putting this innovation into use in real-world applications to reduce the energy costs of computation, especially in the field of artificial intelligence”.

Piergiulio Mannocci, a researcher for DEIB and first author, commented: "This work is the result of an international collaboration between academia and industry that has also involved Peking University, a diverse team involving professors, researchers, PhD Candidates and students, and demonstrates the potential of analogue in-memory computation for high-performance, energy-efficient applications."

The study represents an important step towards more compact, faster and sustainable devices, opening up new perspectives for research and industry. In-memory computing is an ideal solution, in fact, in contexts requiring high performance and high energy efficiency such as artificial intelligence, the processing of large volumes of data, and next-generation wireless communication systems. Applications range from robotics to data centres and from navigation systems to advanced telecommunications networks, such as 5G and the 6G technologies of the future.