Since the 1970's Moore's law stating the number of transistors in a computer doubles every 2 years has remained true. The constant shrinkage of the chip fabrication process to allow for more transistors to reside within the same area has allowed computers to double in speed every 2 years. A question one may ask, concerning Moore's law is, "When will Moore's law fail?". The answer is not simple as it is all relative. If one were to literally translate Moore's Law as the doubling of transistors every 2 years then we are approaching the outer limits of what is physically possible in terms of shrinking transistors. But what if Moore's law were to potentially expand to the field of fiber optics? It is well known that fiber optic cable is significantly faster than copper wiring. Lets look back really quick and go through current computing technologies.
Intel's Ivy Bridge processors from their current Celeron processors to their i7s in the consumer market run on a 22nm node, with the NM abbreviation meaning the transistor size in nano meters. Their previous processors, namely, Sandy Bridge architecture processors ran on a 32nm node. AMD's offerings are their FX processors and APUs (Accelerated Processing Units) currently utilizing a 32nm fabrication process and previously 45nm. Both companies are continuously working to shrink the fabrication sizes within their processors to improve performance.
In 1970 with the Intel 4004, we were at 10 micrometer fabrication, and roughly 10 years ago the fabrication process was 130nm down to 90nm. With the smallest processors currently at 22nm, it truly makes one question when the outer limit will be achieved. The current estimate regarding the outer limit is always changing. The fact of the matter is, when it happens microprocessors will no longer improve in performance, with the exception of improved micro-architectures and better programming practices to better utilize current the CPUs. To at least achieve a fair continuous increase in computing performance, and not necessarily the doubling of transistors stated in Moore's law we will have to find alternative forms of computing.
IBM for the past decade has been researching one of many potential alternatives to conventional silicon computing. Their research involves the introduction of optical based computing. This new technology integrates many conventional circuit components as well as the achievement of optical modulations to drastically increase the bandwidth and performance of computing as we know it. The most practical purposes for this will be to diminish cloud computing bandwidth bottlenecks, but will also make everyday computing activities significantly better as well.
Source, extremetech.com - IBM creates first cheap, commercially viable, electronic-photonic integrated chip
posted by TL
No comments:
Post a Comment