Parallel Distributed Infrastructure for Minimization of Energy

Ten minutes with... Dragomir Milojevic, IMEC

Fri, 2015-05-08

Dragomir Milojevic received his Ph. D. in Electrical Engineering from Université Libre de Bruxelles (ULB), Belgium. In 2004 he joined IMEC where he first worked on multi-processor and Network-on-Chip architectures for low-power multimedia systems. Nowadays, he is working on design methodologies and tools for technology aware design of 3D integrated circuits  as part of the INSITE programme. Dragomir is associate professor at Faculty of Applied Sciences, ULB, where he co-founded Parallel Architectures for Real-Time Systems — PARTS research group. He has authored or co-authored more than 75 journal and conference articles, and served as technical program committee member to several conferences in the field.

 

1. Can you tell me a bit about your main research interests? What led you to work in this field?
My current research interest focuses mainly on the design enablement of future integrated circuits using both advanced device and packaging technologies.  For advanced packing of the circuits we are looking into die and wafer level stacking of the circuits using 3D integration. The objective is to provide means to enable optimal system design using given integration technology.

 

At IMEC we develop process technologies to further enable the benefits of scaling in microelectronics industry. The recent change in the game (scaling wall) forced us to look further the simple CMOS scaling, a model that run for past 50 years but would eventually come to an end. That time is about to come, and we need to find new solutions to enable the extraordinary pace at which microelectronics industry has evolved over the past years. We believe that this is still possible (at least for mid-term developments) if we carefully design systems by co-optimizing the process technology and the system design.

 

2. What areas are you concentrating on within the ParaDIME project?
Two paths lie before us. The first one involves the usage of advanced devices (technologies that use physical dimensions to manufacture transistors that are below 14nm). We are researching how these devices could be used in non-nominal operating points to trade-off the power dissipation of the circuit with the accuracy of computation. 
 
The second path follows the advanced packaging and how this technology could be used to build heterogeneous systems that will save the power by adapting the process technology used to manufacture the ICs with the computation needs. The aim is to combine within the same circuit high-performance, with high-power dissipation CPUs together with low-power, with low-performance CPUs.
 
3. Why is it important for computers to be more energy efficient? What are the major technical challenges which need to be overcome to achieve this?
The overall increase in the computation needs of our society is tremendous and it isn’t likely to change due to the extraordinary progress it provides (despite some computational futilities we witness). On the other hand, energy resources are becoming scarce. Hence, we need to figure out how to enable more computation power -with orders of the magnitude- with lesser  energy required to actually perform these computations. 
 
4. What, for you, are the key technical challenges which need to be tackled in order to achieve more energy-efficient computing systems?
I strongly believe that a lot of progress could be made by much better co-design between the technology, the system and the software. The complexity of the systems, at both software and hardware level led to the adoption of the "divide and conquer" approach: the problem is broken into many smaller sub-problems that are manageable at each one's scale. The problem of such approach is that it necessarily introduces sub-optimality.  So this would be a problem of its own that needs to be solved. 
 
5. Is it possible to deliver genuine energy savings while achieving optimum performance?
Definitively! But I will say this differently: by sacrificing a little bit of performance, important energy savings could be achieved. This could be only a result of a very careful trade-off made across all the levels of the electronic system design: device, system and application.
 
6. Why is the ParaDIME project important? What do you think the most important results will be for society in general?
The ParaDIME project is important because we need, and we will need, more and more computation in the future. And the project is addressing this problem in an original way. 
 
Many of the contributions at one specific level could be applied to other domains as well. Advances in high-performance computing moves forward mobile computing, and vice versa.
 
7. What are your predictions regarding the future of information and technology systems, especially regarding energy consumption and innovative architectures?
In the mid-term, the scaling will continue and we will reach nodes in the range of the few nano-meters. The technology will become extremely expensive but, if carefully used, it could be still cost-effective for some time. We will also have to become more rational and scale only the parts of the systems that need to scale. This could be done using heterogeneous technology integration, and this is where the 3D could help enormously. 
 
In a long-term we need a serious paradigm change. CMOS technology will stop to scale at some point in time; there is no doubt about that (no man can run 100 in 5 sec). We hence need another technology base… optical, quantum computing, or who knows what. This is why this period is extremely interesting for the people working on this field!