Next-gen computing systems inspired by the human brain


  Atienza Alonso David

Data management and interpretation power modern businesses and the global economy. At the heart of this process are scores of data centers that store, manage, and interpret huge volumes of data. However, there is a major scaling mismatch between the proliferation of data centers and the overall power requirements to keep the servers going. Besides, the chances of a system failure increase when computing systems are unable to cool down to an optimum level. To tackle both challenges, Professor David Atienza, director of EPFL’s Embedded Systems Laboratory, has embarked on a project that could potentially create 1000 times more energy efficient systems than those existing today.

Aptly named Compusapien, the project aims to create next-gen computing systems inspired by the human brain. At the core of the research model is a 3D computing server architecture that can obviate excessive power consumption and heat density. Contrary to the conventional 2D setup, the proposed 3D structure envisions a much higher processing speed and reduced latency in big data applications.

Existing computing systems are hampered by the fact that at least a third of the energy intake is frittered away in the cooling process. By introducing better temperature management, the Compusapien model seeks to improve the overall energy efficiency. In this context, Professor Atienza proposes the introduction of tiny microfluidic channels that are barely 50-100 microns high. Sandwiched between layers, the channels have a two-pronged objective: the fluid flowing through them cools the system, while the small fuel cells housed in the channels convert heat into electrical energy. It is proposed to reintroduce this energy in the server system, thus augmenting energy utilization and cutting the energy bill for data centers.

The Compusapien idea has already been tested successfully in the real world. Working with Bruno Michel at IBM Zurich, Dr Atienza demonstrated how 6 watts of electricity (and probably even more) could be produced by implementing the microfluidic technology on a traditional IBM server.

The research highlights the role of well-designed hardware in reducing energy consumption and averting system failures. This augurs well for the future, where the rising demand for technology scaling is likely to exert more pressure on existing systems.

The study is being conducted with a grant from the European Research Council.

Suggested readings: