DCs consume 2% of global energy and will account for 10% by 2030. Their expansion is associated with key energy and climate-related challenges, notably: (a) inefficient energy utilization, (b) high carbon footprint due to their energy consumption and short life-cycle of the servers and (c) impact on the power grid operation.

Indeed, uncontrolled integration of DCs into the power grid affects their operational and environmental cost by: (i) requiring large energy reserves for grid balancing and (ii) large DC’s carbon footprint due to the unaware carbon-content of the used energy. Since the carbon content of the electricity depends on its generation mix, DCs who wish to reduce their carbon footprint, need to optimize their operation with workload scheduling and suitable control strategies. These strategies should account for the CO₂eq content of the electricity generation mix and the use of the DC waste heat.

DCs can be seen as multi-energy systems whose operation can be improved on several dimensions:

1. by producing electricity from organic Rankine cycles (ORC) fed by the DC low entropic waste heat.
2. by local renewable power supply combined with optimal control strategies.
3. by valorizing the waste heat when coupled with local district heating.

At the HeatingBits, a multi-horizon predictive control framework will be developed, and experimentally demonstrated, to operate the EPFL DC and the related campus district energy system with the least carbon footprint and the lowest energy cost. This control framework will leverage advanced power conversion solutions and efficient on-chip cooling for servers.

To increase the DC’s energy efficiency, research efforts will be undertaken in the following areas:

• Heat extraction from DC CPUs of high-temperature through an on-chip cooling system to feed EPFL’s heating system,
• Generation of electricity through a low temperature organic Rankine cycle supplied by the DC recovered heat,
• Efficient integration of the DC power flows within the EPFL energy district thanks to:
  1. a method for DC workload forecasting and control,
  2. a direct current distribution system architecture, and
  3. optimal coordination with the local photovoltaic and ORC power generation, energy storage and heat demand.

Furthermore, the optimized CPU’s cooling and workload scheduling will enable the performance optimization of the servers and extension of their life-cycle, reducing their environmental footprint.

A first-of-its-kind demonstrator on EPFL campus will allow the validation of the developed solutions.

Operate the EPFL DC and the related campus district system with the least carbon footprint and the lowest energy cost.

Showcase solutions of EPFL spin-offs.

Disseminate the technologies developed and proposed solutions through EcoCloud, EPFL’s sustainable cloud computing centre, to major IT companies, such as Microsoft, Intel, HPE, Meta, for joint research commercialization, expanding the project’s impact beyond Switzerland.

The project unites the expertise and competencies of 6 EPFL laboratories in three of its campuses, under the leadership of Prof. Mario Paolone.