Nothing is faster than the speed of light(except the Neutrinos). So you might think that is why HP Labs wants to use photonics (light) instead of electrons to set new computing speed limits. But you would be wrong.
HP Labs' laser-based chip project, codenamed Corona, would use light instead of electricity to communicate information between chip cores and also from the chips to memory in order to save space and to save energy.
 |
| Laser beam |
A Short History of Data on the Wire
If you are old enough (or curious enough), you may have tested your soldering skills assembling electronic bits and pieces into primitive number crunchers consisting of transistors that are, literally, wired together. All those electronic parts and wires were replaced by the integrated circuits, aka microchips, which remain the basis of the modern computer.
An integrated circuit consists, typically, of a silicon wafer that carries thousands, or even millions, of electronic components (resistors, capacitors, and transistors) that are "wired" together by electrodeposited metals. The miniaturization and compression of these components are the subject of "Moore's Law", which observes that the number of components in an integrated circuit doubles every 18 months.
There are two major obstacles foreseeable with the continuing growth in computer processing speeds based on the current technology:
- Miniaturization approaches non-negotiable physical barriers, and
- Energy consumption increases wildly as data processing rates speed up.
Rainbows at Nano-scale
Communicating data by light already occurs on a wide-spread basis, via fiber optics. Scaling laser light generation and integrating it onto microchips lies at the heart of the HP Corona project.
MIT has demonstrated the technology to generate laser light using materials that are compatible with microchip manufacturing processes, meaning that tiny lasers could be built directly into integrated circuits. Once generated, the laser light beams along a nano-scale waveguide many times thinner than a single fiber optic cable.
The trick that really saves space copies a technique called Dense Wavelength Division Multiplexing (DWDM), which is currently used in telecommunications: light of different wavelengths can multiply the information sent over a single "light wire". HP Labs has demonstrated that up to 64 wavelengths can be managed in 64 "ring resonators", a circle along the path of waveguide, in less that one millimeter. Using electrical impulses at 10GHz to "tune" the ring resonators results in 10Gb/s of data transmission times 64 wavelengths for 640 Gigabits of data per second running along a single "wire".
Lightening the Energy Demand
Projecting from the energy consumption of current super-computers, an exascale computer (performing 1018 or a million trillion operations/second) would require its own Hoover Dam worth of power. Most of that power is consumed not in performing operations, but in communicating between parallel processors to schedule tasks and balance loads or in sending data to memory.
Laser-based data transmission reduces heat loads compared to electricity, and the greater bandwidths for communication significantly reduce the power requirements. According to Wired: "Using electronics for a 10-terabytes-per second channel between a CPU and external memory would require 160 watts of power. But HP Labs researchers calculate that using integrated photonics lowers that to 6.4 watts."
Even before exascale computing appears on every desktop, the energy savings could be significant.
Coming Soon to a Desktop Near You
Photonic data transfer between components within an integrated chip needs a decade more research, but data transfer between cores or from cores to memory could come to fruition soon. HP targets bringing a 3-D chip with high-bandwidth photonic communications between 256 processor cores to the market by 2017.
HP's Corona project races against other giants -- such as Intel (Runnemede), MIT (Angstrom), NVIDIA (Echelon), and Sandia National Labs (X-calibur) -- in the search to make high performance computing ubiquitous.
No comments:
Post a Comment