You have probably seen the Back to the Future trilogy. In Back to the Future II Marty McFly and Doc Brown travel to a day in the future. When they made the movie in 1989, they picked a day 25 years out, unimaginably far into the future. But the exact day they picked was October 21 st 2015. Today. There were lots of predictions such as hoverboards (sort of flying skateboards) and self-tying shoelaces that haven’t arrived. One intriguing fact put in the script was the Chicago Cubs winning the World Series. This was obviously put into the script since it was so ridiculous; even in 1989 the Cubs hadn’t won for decades and the drought has now stretched to 107 years. But wait…the Chicago Cubs are in the National League Championship. It could still happen. The semiconductor and EDA industries have their own special aspects of forecasting. Since the 1960s we have had Moore’s Law to guide us. In some ways it has become a self-fulfilling prophecy since the whole industry uses it as a roadmap. It takes the whole industry to move to a new process node since if some aspect, such as tools for manufacturing the next generation don't move, then it is not possible to bring up a volume process in the new node. If EDA doesn’t move, if nobody bothered with double patterning for example, then it wouldn’t be possible to create designs for that node. We have the paradox that we work in the fastest changing industry in the world, with decades of exponential growth behind us, but it is very predictable. However, Moore’s Law has changed in a significant way. It can be looked at as a law about technology, and in that sense it will continue. We have 10nm and 7nm in active development, and test structures being created for 5nm. I have always preferred to look at Moore’s Law as an economic law, which is how Gordon Moore looked at it. What he observed was not so much that process nodes marched inexorably onward, but that the most economically attractive number of transistors to put on a chip did. Put more and the yield fell off; put fewer and you could have got more almost for free due to the planar nature of the manufacturing process. Until recently my favorite way to look at Moore’s Law was that the price of any given functionality halved every couple of years. It was that aspect that meant that a flight simulator that cost tens of millions of dollars in the 1980s has less powerful graphics than a smartphone that costs $500. But it is just that aspect of Moore’s Law that is broken. For now 28nm seems to be a sweet spot, the lowest cost transistor you can buy. Yes, with FinFETs you can get more transistors on a chip, with higher performance and lower power, by using 14nm and 16nm. The latest iPhones and Galaxies use these processes in their application processors. But the cost per transistor is not lower. I don’t think that the general public has really noticed. Even technology journalists still have the attitude that electronics always gets cheaper, and if you don’t like the price point today then just wait. If the application processor in your iPhone goes from $30 to $40 it doesn’t affect the price much. But it does mean that just by waiting the functionality of a modern smartphone is unlikely to be available at a $50 price point. On “Marty McFly Day” here is another prediction from the past. In his famous Electronics article 50 years ago, in addition to the “Moore’s Law” formulation and graph, Gordon Moore said: Integrated circuits will lead to such wonders as home computers, automatic controls for automobiles, and personal portable communications equipment. But the most amazing thing about this statement is not just how prescient it was, but how unlikely it must have seemed at the time. It was made 50 years ago on April 19 th 1965. “Moore Day.” You can read the original article in Electronics here (pdf).![]()
![]()
↧