Experts At The Table: What’s Next?

Last of three parts: Where the value is and who benefits; power improvements in multicore vs. single core and multichip.


Low-Power Design sat down with Leon Stok, EDA director for IBM’s System & Technology Group; Antun Domic, senior vice president and general manager of Synopsys’ Implementaton Group; Prasad Subramaniam, vice president of design technology at eSilicon, and Bernard Murphy, chief technology officer at Atrenta. What follows are excerpts of that conversation.

By Ed Sperling
LPD: Where does the value shift at future nodes? Who makes the money?
Murphy: A lot of the control is moving to the integrated solution vendors. Nokia, for example, has a huge amount of control over TI and ST and Infineon. That doesn’t mean Nokia is making a ton of money, but they’re higher up in the food chain and everyone below is suffering.

LPD: For awhile it was companies like Wal-Mart making the money, not the Nokias of the world.
Murphy: And that’s still the case. Verizon is making more money than the handset makers.
Domic: But in the PC industry, the guys making money were Intel and Microsoft, not HP or Dell. The part that has become very difficult is the EDA industry receives about about 2% of the semiconductor industry’s revenue. The test guys say everyone complains about the cost of test, but they’ve been getting 2% of the semiconductor industry revenue for the past 20 years. The total pie is sliced very fine. If EDA wanted to grow to 3%, someone is going to have to give up 1%.
Murphy: And you can’t just say to the semiconductor guys, ‘You’ve got to give me more money.’ You’ve got to look beyond that, maybe to Wal-Mart.
Domic: But as tasks disaggregate and get more discrete, that makes it harder.
Subramaniam: The reason Intel and Microsoft make more money is they’ve come up with a value proposition that works. If you look at Dell and HP, they’re just assembling things. All the value is in the CPU built by Intel. Whoever builds the next important widget and creates a platform for that widget, they’re the ones who will extract the most value.
Murphy: For a time.
Subramaniam: It’s not clear if there is any equivalent to Intel in the mobile market. Qualcomm is close.

LPD: Is the value moving into the integration of the pieces rather than the pieces themselves?
Domic: I have not seen that. Fabless companies were supposed to do that, but that hasn’t happened. If all a chip company does is assemble pieces of chips you’re buying from different people, there probably isn’t a business model there that works.
Stok: More of the work is shifting there. I’m not sure the value is. The integration is getting more and more complicated. In the end, some people may capitalize on that.
Murphy: Apple is one example of that. They extract more value than other cell phone makers. It’s not based on faster devices or lower power consumption. It’s based on software infrastructure and business infrastructure and branding.
Domic: The car today doesn’t do something much different than it did 30 or 40 years ago. It doesn’t run faster. But it is safer and it has a number of other conveniences. Maybe in electronics we’re going to see an extension from that perspective. My new portable might not be faster but it might have better connectivity. If you drive a car that’s 30 years old, you notice a big difference. There is a lot of extra value that technology contributed. You’ll still be watching images on a TV, but it may be doing more.
Subramaniam: It’s a great analogy but it’s not a pretty one. You probably pay the same today for a car as 30 dollars in real dollars. Twenty years from now electronics will be providing a lot more value, but the consumer will probably be paying less.
Domic: But devices can get a boost not from the chip but from power savings. If you go to Japan and stay in a reasonable hotel, the air conditioning is infinitely more efficient and quieter. The cost of energy is relatively much higher, so it forces economic solutions.

LPD: What you’re talking about is more discipline in terms of where there is no value, and looking at where there is new value. But we don’t always see that value ahead of time. Is that a fair statement?
Murphy: Yes. The whole green industry is suffering from oversupply and under demand. That area was looking promising, but not so promising now.
Domic: But that’s short term. Oil at $140 a barrel will force you into a different scenario than when it is $70 a barrel, but the trend is pointing up. It’s not going to stay.
Stok: It might be the same in the next wave of electronics. As power becomes a real issue, at the next node you can still do a much more efficient design because you make everything significantly smaller. That might lead to a new wave of demand for different devices. There is hope.
Domic: We need something that really increases productivity. A lot of demand has been driven by disposable income. VHS was better than Betamax, but in reality it created nothing new. It’s a nice part of life, but you’re not creating value or extra wealth. The PC created something that, for business, was incredibly convenient and saved money. Applications need a wealth-creation component. That’s why the consumer market is so inconsistent. Years ago AT&T was talking about movies on demand, but those things only shift money. Something like the PC and computers have allowed people to create more wealth.

LPD: Have we hit the maximum number of cores or will there be more?
Domic: We have not hit the limit. For EDA we have been forced to make things more parallelizable. You will see more cores going forward. If you think of an airline reservation system, they are doing millions of small transactions.
Subramaniam: There are a lot of applications that can take advantage of multiprocessing. Medical research, where they study the interactions of molecules and proteins, requires a lot of computation at a minute level that can be parallelized. But for traditional applications on a PC, it’s not clear you can benefit from many cores.
Murphy: You can always throw as many cores as you want at verification.
Domic: In EDA, there have been announcements from Synopsys and Cadence in layout to take advantage of new cores.
Stok: We already have systems with many cores. They just have many chips. You’ll see those cores move onto single chips because you get power efficiency.
Domic: We benefitted in EDA in the past if you went to the next generation. You got 2x performance. Now we don’t get that performance increase. We have to rewrite pieces to give you that performance.