SAN JOSE, Calif. -- Some of the world's smartest semiconductor engineers will gather at Semicon West to discuss ways to keep Moore's Law alive. Eventually we will not be able to keep making smaller, faster cheaper chips -- at least not using anything like the techniques we use today.
Many stories will be written about the death of Moore's Law shortly, given the wealth of sessions about the silicon road map at Semicon West. I'll probably write at least one myself. Although I have no idea what I will say yet, I suspect it might take the form of an upbeat look at the latest batch of promising ideas from world-class engineers.
I hope to hear discussions at the event about the outlook for a 5nm node, extreme ultraviolet lithography, 450 mm wafers, and more. Some of the world's top experts on making chips will be hanging out in San Francisco (where more than a few of them live anyway). I hope to meet as many as possible.
I wrote my first story about the death of Moore's Law back in the mid-1990s, when I had my first and only interview with Gordon Moore. He said his observation about doubling transistors in a given area every 18 to 24 months was bound to end -- it will hit atomic limits. Before it ends, it will get really hard to do and slow down, he added.
Many experts think we're seeing that now. In May 2013, Broadcom co-founder Henry Samueli was the first smart semiconductor executive I heard admit publicly this is starting to happen now.
The need for using two or more passes through a 193nm lithography machine to make 20nm and smaller circuits is one of the factors some point to when they say Moore's Law is already dead, as Zvi Or-Bach said earlier this year. They say the extra time on the litho machines and the extra masks required mean the next nodes will not reduce the costs per transistor, one of the three corollaries of Moore's Law -- that transistors will get smaller, cheaper, and faster with each generation.
Today, some say the situation is more like ordering dinner at a Chinese restaurant. You pick one or two items, but not all three.
That's a matter of some debate. In a recent story, Chris Bencher, the litho guru at Applied Materials, noted NAND flash has been using double patterning since 2008 and is starting to use quad patterning -- and those flash cards at the electronics shops just keep getting cheaper. But even Bencher noted the same will not hold true for logic chips which have unique requirements that will drive costs up.
Next page: Violent reactions