Cool Chips 2009 (COOL Chips XII)
IEEE Symposium on Low-Power and High-Speed Chips
Yokohama, Japan
April 15-17, 2009
COOL Chips is an international symposium, started in 1998, for new low-power and high-speed chips, with a focus on microprocessors and their applications. The twelfth edition, COOL Chips XII, was held in Yokohama on April 15 to 17, 2009, at the Yokohama Joho Bunka Center (Yokohama Media and Communications Center). This post was first published on March 25, 2009 as an event notice and is kept as a dated record.
What the 2009 symposium covered
The organizers' call for contributions (now available only through the Internet Archive) asked for papers on the architecture, design and implementation of chips in three broad areas:
- Low-power, high-performance processors for multimedia, consumer electronics, mobile, graphics, encryption, robotics, networking and biometrics.
- New architectures for single-core and multi-core designs, embedded systems, reconfigurable computing, grid, ubiquitous and dependable computing, and wireless.
- "Cool" software, meaning binary translation, compilers and low-power software techniques.
The 2009 call also had a tenth-anniversary special topic on low-power, high-performance supercomputing. The organizing committee planned to ask IEEE Micro to publish selected papers in a special issue on COOL Chips XII.
Why a "cool chips" conference matters
Power sets hard limits on processor design. A chip can only burn as much power as its package, battery and cooling allow, so every gain in speed has to be paid for with better energy efficiency. COOL Chips brought together the people working on that trade-off: processor architects, circuit designers and compiler writers.
The 2009 program shows three themes that were big at the time:
- Many-core processors, where many simple cores share the work instead of one very fast core.
- 3D integration, stacking dies on top of each other and connecting them with through-silicon vias (TSVs), vertical copper connections that pass through the silicon. Short vertical links use less energy than long wires across a large die.
- Reconfigurable processors, which change their hardware datapath to match the task, aiming at the efficiency of fixed hardware with some of the flexibility of software.
Deep dive: the numbers and details
This part holds the program details, the figures quoted by the speakers and a worked example on 3D yield. Skip it if the overview is all you need.
Keynotes and special sessions
The COOL Chips XII advance program (Internet Archive copy, March 2009) listed:
| Session | Speaker | Topic |
|---|---|---|
| Keynote | Hiroshi Ishiguro, Osaka University and ATR | Future robotics and the chips that robots working in homes and towns will need |
| Keynote | Cisco Systems (ERBU engineering director) | Cisco QFP, the Quantum Flow Processor networking chipset |
| Keynote | Mitsumasa Koyanagi, Tohoku University | New 3-D integration technology and 3-D systems |
| Special session | Yuan Xie, Pennsylvania State University | Designing cool chips using 3D stacking technology |
| Special session | Hideharu Amano, Keio University | Low-power techniques for reconfigurable devices, focusing on dynamically reconfigurable processors |
Figures quoted in the abstracts
- Cisco QFP: described as a fully integrated, programmable networking chipset with 40 cores and 800 million transistors, able to run up to 160 processes at the same time. It powered the data plane of the Cisco ASR 1000 series routers.
- Reconfigurable processors: Amano's abstract states that dynamically reconfigurable processors achieve 8 to 10 times lower energy than DSPs in comparable technology. The session reviewed operand isolation, reducing reconfiguration, Vdd control, dual Vdd, dual Vth and power gating.
- 3D integration: Koyanagi's abstract notes that in wafer-to-wafer stacking the overall chip yield falls exponentially with the number of stacked layers. His group proposed "super-chip integration", in which more than a thousand known good dies are aligned and bonded at once by self-assembly.
Worked example: why 3D stacking needs known good dies
The exponential yield loss is easy to see with numbers. Illustrative figures only: if each wafer layer has a die yield of 90% and dies are stacked wafer-to-wafer without testing, a stack is good only if every layer's die is good.
| Layers stacked | Stack yield at 90% per layer | Stack yield at 80% per layer |
|---|---|---|
| 2 | 81% | 64% |
| 4 | 66% | 41% |
| 8 | 43% | 17% |
Yield is Y to the power n for n layers. Testing dies first and stacking only known good dies (die-to-wafer or die-to-die) breaks that compounding, which is the motivation behind the self-assembly approach in the keynote.
Author schedule
The call for contributions set an extended abstract deadline of January 26, 2009, acceptance notices on March 2, 2009 and final manuscripts on March 16, 2009. Poster abstracts, one page in two-column IEEE format, were due by March 9, 2009.
What happened next
The series has continued. The call for the 29th edition, COOL Chips 29, set it for April 15 to 17, 2026 in Tokyo, under the slightly wider name "IEEE Symposium on Low-Power and High-Speed Chips and Systems". Its topics show how the field moved: AI accelerators from cloud to edge, high-bandwidth and compute-in-memory designs, photonic computing, dense 3D architectures and hardware security.

Your comments will be moderated before it appears here.