High-performance computing is required when solving some of the world’s more advanced technical problems. Unfortunately, the amount of energy required to power the computer clusters and cool them is growing dramatically every year.
Assistant Professor Ayse Coskun (ECE), Jie Meng (PhD ’13), and Daniel Rossell (MS ’12) are studying this problem and looking at how 3D-stacked multicore systems could allow for the integration of DRAM layers and processor cores on the same chip. This would not only improve memory bandwidth but also reduce memory access time, a long-standing performance bottleneck in computing.
“There’s a big push right now to make chip sizes smaller and more efficient,” said Rossell. “Using novel design techniques can actually improve performance without increasing the energy required.”
The team recently shared their findings at the High Performance Embedded Computing (HPEC) Workshop, held at MIT Lincoln Laboratory, and their research paper, “3D Systems with On-Chip DRAM for Enabling Low-Power High-Performance Computing,” won the Best Paper Award.
Coskun believes that part of the reason their research stood out was the team’s ability to support their thesis with substantial data.
“Using 3D systems to improve performance is a relatively new idea,” said Coskun. “Much of this research is in early stages, but we were able to provide detailed information on performance, temperature and energy simultaneously.”
In fact, Coskun, Meng, and Rossell found that by using a 3D system, application performance improves by 72.6% on average when compared to a conventional chip with off-chip memory.
Rossell, who became interested in computing research after taking a class with Coskun, said that Coskun has been great to work with.
“She’s very easy to get along with and always gives us just enough information to get the ball rolling on our research but not so much that we aren’t learning,” said Rossell.
Prior to working with Coskun, Meng was new to design automation and computer architecture research. Now, Meng’s work has led to a paper that was presented at the Great Lakes Symposium on VLSI in May 2011, and she also co-authored a journal article that was published in IEEE Micro. Last summer, Meng worked at Sandia National Laboratories, where she pursued research ideas on improving the performance and energy efficiency of multicore systems.
“Professor Coskun has been very helpful over the last two years and always offers good insight,” said Meng. “Working in PeacLab has been very productive and fun.”
For more information, visit the Performance and Energy-Aware Computing Laboratory (PeacLab) website.
-Rachel Harrington (firstname.lastname@example.org)