Project description
Satisfying big data centres hunger for more power
More data, more power. The power consumed by data centres is likely to triple in the next decade. Poor energy efficiency is due to the physical separation between the central processing unit (CPU), where data are computed, and the memory, where data are stored. There is a way to solve this. A new paradigm has been already developed to execute machine learning (ML) tasks in just one step within memory. In this context, the ERC-funded CIRCUS project will build on this technology to improve its scalability and technical feasibility by simulations and realisation of a small-scale prototype. The project will also conduct a market search and draft an investor-ready business plan.
Objective
Every second, our smart phones deliver a wealth of information that can be used to monitor the traffic, the financial transactions, and even the spread of a dangerous disease. The processing of these big data into a meaningful information requires specific machine learning (ML) algorithms, which essentially consist of regression techniques for inference, classification and prediction. The conventional digital computers are not designed to optimally solve these problems with efficient time and energy consumption, which is one of the reasons why the power consumption by data centers worldwide is expected to triple in the next decade. Such a poor energy efficiency is essentially due to the physical separation between the central processing unit (CPU), where data are computed, and the memory, where data are stored, according to classical von Neumann computer architecture. In the frame of our ERC-CoG RESCUE, my group has developed a new paradigm to efficiently execute ML tasks in just one step within the memory. Instead of moving data from the memory to the digital CPU, an analogue computation is directly operated within the data, thus breaking all previous limits of time and energy consumption (10.000x reduction in the number of operations, hence time, and 1.000x in energy). Our in-memory technology is modular and universal, thus can be implemented in any existing memory and computing technology to accelerate ML tasks in future smartphones and data centers. In the ERC-PoC CIRCUS, we aim at bringing this technology to a higher maturity level, demonstrating its scalability and technical feasibility by simulations and realization of a small-scale prototype. In the meantime, we will also perform a comprehensive market search to recognize opportunities and draft an investor-ready business plan for raising future investments to further advance the solution toward industrial exploitation.
Fields of science
- engineering and technologyelectrical engineering, electronic engineering, information engineeringelectronic engineeringcomputer hardwarecomputer processors
- natural sciencescomputer and information sciencesdata sciencebig data
- engineering and technologyelectrical engineering, electronic engineering, information engineeringinformation engineeringtelecommunicationsmobile phones
- natural sciencescomputer and information sciencesartificial intelligencemachine learning
Programme(s)
Funding Scheme
ERC-POC - Proof of Concept GrantHost institution
20133 Milano
Italy