It may not be quite as exciting as a photonic transistor, but IBM’s latest technology breakthrough still raises hopes of a tremendous increase in computing power owing to photonic technology. The company’s new technology, dubbed CMOS Integrated Silicon Nanophotonics, integrates electronic and photonic devices onto the same silicon chip, potentially allowing faster and higher-performance connections between racks, servers, chips, and even devices on the chip.
Although the new technology does not have any immediate implications for increasing the speed of a processor, it has tremendous promise in the area of supercomputing, as well as in more mundane computers and computer networks. According to PhysOrg.com (“IBM's breakthrough chip technology lights the path to exascale computing”), the technology uses “the front-end of a standard CMOS manufacturing line and requires no new or special tooling.” As a result, the technology has greater potential for commercial success right from the start.
The technology permits greater integration of photonic devices, allowing the use of light pulses for communication among servers, devices, or chips. This technology, if IBM is able to flesh it out into a viable device or product, could help the company meet its goal of constructing an “exascale” computer by 2020. An exascale computer would be able to perform on the level of exaflops: 1018 computations per second (that’s a million trillion—or a billion billion—computations per second). Current supercomputers operate in the petaflop (1015 computations per second) range. Thus, IBM is hoping that its new nanophotonics technology will yield a 1,000-fold increase in computing power.
Again, this technology doesn’t increase processor speed. But processor speed is certainly not the only factor when it comes to computing power. In large-scale computing especially, individual processors are limited in performance by delays in, for instance, retrieving data from storage or even caches. These delays drive the need for each device (processor, cache, storage device, or other device) to be in close proximity to other devices. Furthermore, the speed of light places a fundamental limit on how quickly information can be transmitted across a given distance. Optical interconnections among devices, however, are in many ways superior to copper interconnections. Fiber optics is a widely used method of transmitting large amounts of information over long distances in a relatively inexpensive and reliable manner.
Applying a similar principle on a smaller scale—in particular, on the scale of a single chip—has yet to be realized, however. IBM is hoping that its nanophotonics technology will be the key to unlocking this new realm of possibilities. By speeding communication between devices and components on a chip, some of the limitation placed on processors by the (lack of) proximity to partner devices is removed. Of course, the fundamental limit remains—transmitting data over one meter, for instance, will always take at least 3.33 nanoseconds (at least if you believe the theory of special relativity). But IBM’s new technology may help bring current interconnects closer to this limit.
Although IBM’s focus for its new technology is limited to interchip connections, it is not discounting the possibility of reaching an even finer level of communication. According to Computerworld (“IBM chip breakthrough may lead to exascale supercomputers”), IBM photonics research scientist Will Green expressed hopes for enhanced intrachip connections, but the technology is not yet developed to that point: “There is a vision for the chip level, but that is not what we are claiming today.”
For supercomputing, the implications of a successful application of IBM’s technology are clear: greatly expanded potential computing power. Data centers may also benefit, although the benefits may be less pronounced. Nevertheless, the integration of photonic devices directly on silicon chips—and with no special new tooling technology beyond standard CMOS—could yield improved computing power. Furthermore, such integration may also open up possibilities for other technologies as more devices of different kinds become able to coexist on individual chips. As the DCJ has noted with other newly announced technologies, the proof really is in the pudding. Although the new IBM photonics technology sounds great with its promises of greatly increasing computing power, it’s only as good as a working device that employs it. Although development of a new technology is admirable, application of that technology to produce a result that yields a real and measurable performance improvement is even more admirable. In the meantime, we can only wait and see what IBM is able to do with this technology.
Read more about chips by jeffrey clark
No comments:
Post a Comment