In developing an application to integrate 128 processors on a single chip, ST Microelectronics turned to Leti for assistance. To take on this task, Leti called in reinforcements, organizing and launching the European PRO3D Consortium, with ultimate goal of developing a holistic approach to system design that encompasses software, architecture and 3D integration. 3D Incites spoke with Ahmed Jerraya, Leti’s research director and head of strategic design programs, about the consortium’s goals, and the roadblocks it aims to conquer.
The Driving Force
Multicore (or manycore) processors are not a novel concept for IC design – the trend began a few years back. In fact, most of today’s PCs and laptops include a CPU that may be dual or quad core. In addition, it also include a graphics processor that may include dozens to few hundreds of computation units (small processors). Jerraya explained that these designs are based on 2D architecture with processors, memory, and logic interconnected side-by-side on the same big chip. But now, due to all the classical reasons to go 3D, plus one essential driving force, manycore designers are looking to 3D architectures for solutions. That driving force is known as the Memory Wall. “The bottleneck is the link between the processor and the memory; The’more powerful the processor, the bigger the link to the memory, explained Jerraya. That’s where 3D integration comes in, stacking memory on the processor allows for increased interconnect density and a shorter distance. In concept, it seems simple but the solution is complicated, and according to Jerraya, requires a holistic solution in which hardware and software are developed simultaneously.
Thus the impetus for Leti to launch the European PRO3D Consortium. Leti recently announced the six organizations who will participate in the consortium. Collaborating with ST Microelectronics and CEA-Leti, are: Verimag, Joseph Fourier University, Grenoble 1, France; ETHZ, Swiss Federal Institute of Technology Zurich; Alma Mater Studorium, University of Bologna, Italy; and EPFL, Federal Polytechnic School of Lausanne, Switzerland.
The Master Plan
While some technology issues will be addressed by the consortium, Jerraya said it will mainly focus on two software-related projects: developing a 3D architecture exploration environment; and development of a software environment for 3D cooling. The former requires simulation of whole system to define next generation; the latter addresses software that exists as part of the 3D system to specifically manage thermal dissipation. For example, consider a technology with micro-channels that contain liquid. The speed of the liquid is controlled by software that alters cooling according to heat generated by the 3D architecture.
“Each of the partners in the consortium has existing technology to cover the food chain. The idea is to align all the technologies to create a holistic solution,” said Jerraya, adding that the project will run for 3 years, with 10 people dedicated to it for a total time commitment of 30 man-years. While there is no common lab per se, the members meet weekly and joint work is well synchronized in terms of milestones and deliverables. Leti is focusing on two areas; providing 3D TSV technology and also working on hardware/software integration. ST Microelectronics is providing the infrastructure for the project. ETHZ will lead the architecture exploration, developing tools to compare different possibilities. EPFL will work on the active cooling software for 3D architectures. The University of Bologna will focus on developing soft ware for network architecture. Verimag will be responsible for providing software verification tools that ensure the timing of the on-chip software is correct.
Summing it up
Here’s the gist of it all. In defining a new 3D architecture, it’s necessary to develop software in several layers. To achieve this, it’s critical to simultaneously develop the architecture and software. Today’s multicore processors may be fast, but the full potential is not being realized due to the individual core’s lack of ability to “talk” to each other. Programming the cores to work together as part of the development processes will essentially remove this block. Remove the blocks, and the Memory Wall comes tumbling down. The European PRO3D consortium is on the case. Stay tuned for future progress on 3D InCites.
