Hutcheson, who is CEO of VLSI Research in Santa Clara, Calif., has seen scores of major factories around the world. He has a detailed understanding of the industry supply chain and equipment manufacturers, and has a record of success predicting trends based on sophisticated economic models. Amid swirling debate over the strengths of fabs versus fabless chipmakers, shrinking geometries and node transitions, Hutcheson shared his perspective on the industry today, the future of Moore’s Law and the transition to 450-milimeter wafers.Dan Hutcheson has been following the semiconductor industry for more than 30 years and ranks among the foremost independent authorities on chip making and the economics of innovation. In the early 1980s, he developed the first factory cost-of-ownership models, and more recently advised the White House Council of Economic Advisors on innovation.
How do you see the industry today?
We’ve gone through 2 decades when foundries were growing faster than the industry. Then the fabless industry was growing faster than the industry, and the IDMs [integrated device manufacturers] were going to go away. Maybe 8 years ago, I predicted that as we moved below a hundred nanometers, design and manufacturing were going to become interated again, that the key strategic thing that made the foundries possible was going away and that you needed a lot tighter coupling between manufacturing and design. You’re seeing that now.
It used to be that the foundries were neck-and-neck with the IDMs on process technology and now they could be a generation behind. Certainly a generation behind Intel in the logic space. They’re also behind the memory IDMs. Intel is already racking 22-nanometer, and they’re still trying to get to entitled deals on 28 nanometers.
If you look 5 years out, how will the market landscape for fabs, foundries and process tech be different than it is today?
If you look at the foundry industry, there’s only one company in that world that has been consistently profitable. I think what you’re going to see is more consolidation in the foundries. It’s going to narrow down to one, and they will become more collaborative with their fabless suppliers.
We’ve already seen that the foundries’ share of the market is not changing in terms of share of non-memory. In the mobile communications chip market everybody knows Qualcomm is No. 1, and you say, “Well, who’s No. 2?” Most people don’t know it’s Intel. There’s this belief that Intel can’t compete in mobile, and yet they’re No. 2 in the space already.
As you put more and more transistors on a chip, and you get that ability to put a billion transistors on a chip, you’ve got the entire system there. So now you’re putting all this other system architecture onto the chip itself. That means that the whole surface of the chip is very different.
How are the manufacturing and process worlds changing as we scale to geometries, materials and manufacturing techniques that are increasingly complex and cost prohibitive?
In the ’90s and the 2000s, if you followed the design rules and the design guidelines, you didn’t have to worry about manufacturing; you could just make your design, and they pretty much worked coming right out of the box.
Now, design can be perfect, but if you haven’t dealt with all the lithography hotspots, all the various issues of the actual manufacturing of a chip, because everything is so close together, it’s very difficult.
You mentioned complexity and the challenges they present for manufacturing. How important is process technology in today’s competitive environment?
It’s critical because of the complex supply chain between companies that are foundries and companies that are fabless — there’s inter-competitive issues.
One of the advantages foundries had in the ’90s and the 2000s was that it was cheaper to produce in Taiwan. Labor and engineering were a higher percentage of the total cost, and many IDMs were, frankly, inefficient in the way they ran their fabs. Today, the IDM has a huge advantage because there’s no sales, marketing or purchasing interface between the designers and the process engineers. They both can work together and solve problems, and this is why they’re able to do it faster. They have shorter learning loops. They have shorter decision cycles, and that means they get answers faster.
The other thing is the fabless company doesn’t necessarily want to give the design to the foundry. The IP is pretty leaky in the fabless/foundry space and that means that the designers don’t necessarily want to give a leading-edge design to the foundry. So the foundry runs their process off of SRAMs and test chips and stuff like that.
Some say it’s yield related, some say it’s investment related. What do you make of the 28nm capacity shortfall we’ve seen in recent headlines?
If you have a yield problem, you also have a capacity problem. In reality, it was a demand problem; there wasn’t enough demand for 28 nanometer. There wasn’t enough demand because the yields were so poor. Now it’s a capacity problem because they’ve solved most of the yield problems, so now there’s this huge flood of demand for 28 nanometer.
Looking ahead, how will the rate of materials innovation change as we get to smaller nodes like 14nm and 10nm?
I think the interesting thing about materials is that we’ve gone through an era of massive materials innovation to deal with the limitations of devices. Tri-Gate resets the whole game because now you’re back to transistor physics, and it’s not a materials game. It’s a structural game, and you’re going to a much better structure that gives you a lot better device performance for lower power consumption. It really is — I hate to use the word — a game-changer. You’ve seen almost everyone in the last year realign toward a thin-set type of device. The huge breakthrough that Tri-Gate represents in terms of technology. It’s like going from prop engines to jet engines.
Why do you think Tri Gate is a game-changer?
It opens up the world of semiconductors into the sub-20nm future, and getting down to 10-[nm] and sub-10nm. The biggest advantage is that it’s a much more efficient device. I could say, “It’s like going from a gas-engine car to a hybrid” in that it consumes a lot less power, but the bad thing about that example is that hybrids aren’t known for their performance. So it’s like going from a 4-cylinder to a V-12, and getting all the advantages of a hybrid at the same time. It’s a unique technology because these things just don’t come along very often in history.
Do the challenges in sub-micron mean we’re nearing the end of Moore’s Law?
We used to measure everything in microns and now we’re measuring everything in nanometers. That was when you started seeing so many companies start to fall behind because there were just so many technical issues. There was a time when I could come to the equipment suppliers and get a total solution and actually be in semiconductor manufacturing. I could buy my design tools, get my IP and it all worked together. Once we crossed that [nanometer] barrier, we ran into all kinds of problems.
So far, Moore’s Law is continuing to go on. What makes Moore’s Law so powerful is that he [Intel co-founder and former CEO Gordon Moore] describes that you get twice the components every year. It was every year back then . In ’75 he changed it to every 2 years, but you get twice the components per period of time, and it costs you roughly the same to get 2x the components. That’s what drove the industry, because it meant that you would always get better and better electronics and it wouldn’t cost you more. What we saw was the cost of computing just went down orders of magnitude.
The interesting thing that’s happening, though, is people still want more density, they want more performance. On their cell phone, they want it to do more computing tasks, they want video, they want streaming, and they want the battery to last for a day, maybe 2. That means that you still have to strengthen transistors, and you still have to put more density into them. We’re moving into a new world that wasn’t in the original papers that Moore wrote. We’re moving from density drives cost to density drives ability to pack more transistors into a given area and have a lower power consumption. So I get more performance for the same power. Whether we call it Moore’s Law or something else, it’s going to continue to drive this industry for at least another 10 years.
I tend to believe that Moore’s Law is an article of faith. If you try to apply a scientific method to predicting the end of Moore’s Law, you can always show what you can’t do in science. And the problem that you can’t show is how you are going to get around that. Who’s going to come up with the next great idea? And usually, it’s one or two people on the whole planet that come up with the next great idea.
Look at the tablet phenomenon. People have played around with tablets and they went nowhere for 20 years. Then Apple cracks the code and they introduce the iPad. It takes off and it’s created huge new markets for semiconductors. People say, “Oh, it’s killing the PC,” right? Well the fact is, the PC has continued to grow, the PC industry has continued to grow … even [for] Apple! Apple’s selling far more Macs today than they sold before they introduced the tablet.
Nvidia, TSMC and others have called for the industry to move more aggressively to 450-millimeter wafers. What’s driving the speed of this transition?
The timeline we see is really toward the end of this decade. There’s just so much that has to happen; it’s not like you blink an eye and you’re at 450mm wafers. Even back in the days where everybody made their own tools — like Fairchild, they made transitions maybe, once every year or two in wafer size. The 300mm wafer took a good decade to bring to manufacturing. We’ve seen that time stretch out because there’s so much technical work, and there’s also a lot of investment that has to be made because the silicon manufacturers have to invest in the crystal-pulling furnaces and all the technology it takes to make a 450mm wafer and polish it.