It was a surprise to me that so much of transistor performance improvement can be attributed directly to improvements in materials. Another example of how chip makers are going to enormous length to squeeze out every drop of performance.
In chip design, your problem is particle plus yield. That is not a circuit problem under most circumstances. It also makes the case why inside foundry could help drive innovation since you can act faster to fix material and production issues compared to outsourcing manufacturing.
Well, if you don't look at the system level (where comparison is always tricky at best, just look at the current dispute "Intel beating ARM") people in chip manufacturing rate the technology in terms of plain transistor data like Ion/Ioff ratio, Gate leakage etc. Pretty much all the recent improvements there resulting in performance vs power gains came from material changes reducing leakage and improving carrier mobility. I would say 90% may sound a bit on the high side but looking at actual data for 45/40, 32/28 nm technologies this is quite true.
The design and system level aspect comes on top, otherwise Intels 22 nm products would plainly rule the world. Still they don't...
Yes, that 90 percent claim does beg a bit of clarification. Materials have always played a huge role in boosting performance...why is that so different now? Isn't size still important, or is that now credited to materials?
The basic material is still the same (silicon)...if it was something dramatically different like graphen that would be another story...so what is new are some aspects of silicon engineering, strained channels, use of SiGe, new epi layers, 3D transistor nature etc
ON re-reading this story, it sounds like the 90 percent figure stems from the way materials are applied, not from the materials themselves. It may sound like a slight difference, and I'm still a bit dubious, but here's what the story says (emphasis added):
Today, there are several epi steps in the semiconductor manufacturing process, Chu said. More are on the way, he told us, as chip vendors see the potential to improve the performance of their devices using this technology. Applied defines epitaxy as a method of depositing or growing a monocrystalline film where the deposited film takes on a lattice structure and orientation identical to those of the substrate.
I checked back with an Applied spokesperon on the 90 percent claim. This was her response:
"Applied works closely with customers to improve device performance and they calculate that up to 90% of the performance gains they are seeing today are being driven by new materials innovations and new device architectures compared to lithographic scaling."
Still sounds like a sales pitch to me, but that's about what I'd expect from a marketing person for the company who's selling the stuff. I still think it would be more credible coming from an outside ratings agency -- if such a thing exists for that.
Blog Doing Math in FPGAs Tom Burke 0 comments For a recent project, I explored doing "real" (that is, non-integer) math on a Spartan 3 FPGA. FPGAs, by their nature, do integer math. That is, there's no floating-point ...