Archive for the ‘ interconnect ’ Category

IBM Nanophotonic Switch Promises Faster Energy-Efficient Computing – 2010-03-04 13:41:03 | Semiconductor International

IBM Nanophotonic Switch Promises Faster Energy-Efficient Computing – 2010-03-04 13:41:03 | Semiconductor International.

Alexander E. Braun, Senior Editor — Semiconductor International, 3/4/2010

IBM scientists in Yorktown Heights, N.Y., have taken a significant step toward replacing electrical signals that communicate via copper wires between computer chips with silicon circuits that communicate using pulses of light.

The device, called a nanophotonic avalanche photodetector (NAP), is the fastest and smallest switch for directing traffic in on-chip optical communications, ensuring that optical messages are efficiently routed. It could enable breakthroughs in energy-efficient computing with significant implications for the future of electronics.

IBM Nano detectorNanophotonic germanium photodetector generating an electron and holes avalanche. (Source: IBM)The NAP exploits the “avalanche effect” in germanium. Analogous to a snow avalanche on a steep mountain slope, an incoming light pulse initially frees just a few charge carriers, which in turn free others until the original signal is amplified many times. Conventional avalanche photodetectors are unable to detect fast optical signals because the avalanche builds slowly.

“This invention brings the vision of on-chip optical interconnections much closer to reality,” said T.C. Chen, a vice president at IBM Research. “With optical communications embedded into the processor chips, the prospect of building power-efficient computer systems with performance at the exaflop level might not be very distant.”

IBM’s avalanche photodetector can receive optical information signals at 40 Gb/sec and simultaneously multiply them tenfold. Since the NAP operates with just a 1.5 V voltage supply — 20× smaller than previous demonstrations — many of these tiny communication devices could potentially be powered by just a regular AA-size battery, while traditional avalanche photodetectors require 20-30 V power supplies.

“This dramatic improvement in performance is the result of manipulating the optical and electrical properties at the scale of just a few tens of atoms to achieve performance well beyond accepted boundaries,” said Solomon Assefa, one of the principal researchers. “These tiny devices can detect very weak pulses of light and amplify them with unprecedented bandwidth and minimal addition of unwanted noise.”

In the NAP, avalanche multiplication takes place within a few tens of nanometers and happens very fast. The small size also means that multiplication noise is suppressed by 50-70% with respect to conventional avalanche photodetectors. The device is made of silicon and germanium, materials already widely used in production of microprocessor chips. Moreover, it is fabricated using standard semiconductor manufacturing processes; thus, thousands of these devices could be built side-by-side with silicon transistors for high-bandwidth on-chip optical communications.

IBM has worked on this for the past few years, according to Assefa. “We have developed what you might call a nanophotonic tool kit,” he said. “We have made most of the devices that we need, such as modulators to modulate the light, waveguides, switches and all the other components to build on-chip interconnects. The NAP is the last piece of the puzzle, which we needed to have one chip send encoded pulses of light, and the next chip receive it and distribute it. Now the next step is to continue the development and production of these nanophotonic devices along thin-film transistors. If we put all of this together, we believe that within 10-15 years we will be able to integrate onto the chips with the microprocessors, this photonic interconnect system for networks.”

Assefa indicated that microprocessors are becoming extremely powerful, with several increasingly advanced cores being packed in. “This has great potential; however, the next step to make these computers even more powerful is to enable them to communicate. For this, high efficiency and low power usage are important to send as much data as possible from one chip to another.”

With copper, it would be very difficult to send this tremendous amount of data because considerable more power would be required and the excessive heat created would have to be dissipated. “If copper is replaced by light pulses, considerable power is saved and one can send a hundredfold more data than is possible with copper wires,” Assefa said. “At present, we are talking about communication between chips. In the future, similar devices could be used for communication between the different cores. In principle, you can think about a cell phone-sized computer as powerful as today’s supercomputers. It is in these kinds of applications that silicon nanophotonics will have a considerable effect.”

Assefa said he views this as more than just a scientific breakthrough — it is a nanophotonics paradigm shift. “There are many nanoscale effects that we could harness for our benefit,” he said. “In the NAP, for example, we can confine extremely high electric fields over a few tens of atoms. This is a new realm of physics. There is much that can be done to improve device performance. The 1.5 V low-power requirement is a direct consequence of this nanoengineering. If we can continue to leverage these nanoscale effects, many more advances will be made. There are no additional costs, no new processes that must be introduced. There are only advantages.