posted by Bryon Moyer
With semiconductors, we have this expectation that, at some future time, we’ll be able to integrate everything onto a single chip. Analog, digital, MEMS… you name it.
And, theoretically, we will be able to. In fact, we probably can now. Do we? Nope. And it’s even less likely as we keep moving forward.
Why? Because the most advanced wafers are freakin’ expensive. If there’s a bunch of your stuff that will work at 28 or 180 nm, you’re going to do that because it’s so much cheaper than, say, 14 nm. And if the different components have different yields, then you don’t want to be throwing away large chunks of expensive good silicon because one small area is bad. So you’ll use the cheapest possible process for each element, and then bring the individual tested chips together on an interposer of some sort – or maybe even stack 3D – and call it good.
Yield makes a big difference here. If you think about what InvenSense does with their gyroscopes, for example, they take a MEMS wafer and an ASIC waver and mate them face-to-face. That means that you may end up throwing away a good MEMS die if it happens to mate with a failing ASIC die – and vice versa. So the only way this works is if the yield is high enough to make such loss negligible. If that’s not the case, then you want to test and singulate the wafers independently and co-package only the good units.
So… that’s semiconductors, over there in that corner. Over here in this other corner, we have printed electronics. Whole different ballgame. And it brings with it the promise of printing entire systems in a single roll-to-roll printing pass.
Or… maybe not.
I had a conversation with Thinfilm at last November’s IDTechEx show. They make a wide variety of memories and other components printed on plastic. Their first generation was roll-to-roll memory for “consumables” – things that will be used and then discarded, like labels*. And it was printed on rolls 12” x 1 km.
The second generation, however, has more than just memory. There may be sensors and radios, such as are on their sensor labels.
But these are no longer printed in a single pass on a long roll. In fact, some of it isn’t even roll-to-roll. And some of the components, at present (like passives), aren’t even printed; they’re discrete.
Why not just do them all together? Same reason as we don’t with semiconductors: yield. If you commit everything together, then you may end up throwing away lots of good stuff due to a little bit of bad stuff. So their printed subsystems are manufactured and tested – in three different locations (Korea, Sweden, and San Jose), and then these are mounted on a plastic substrate with printed connections. This last step is a sheet process, not roll-to-roll.
Might this eventually evolve, as yields improve, to higher levels of integration? Yes, they say. But there’s always going to be a leading edge, and leading edges tend not to yield as well as established processes. So any product incorporating aggressive, novel technology is likely to be pieced together.
In other words, we’re unlikely ever to be working solely on a fully-integrated roll-to-roll basis.
You can find out more about Thinfilm here.
Meanwhile, there’s whole different reason for resisting integration that we’ll talk about in a few days…
*The military also uses the word “consumable” as a quaint euphemism for ammunition – hardware that will be used once and then be, shall we say, taken out of action. And “smart” versions of such hardware will have electronics on them.
(Image courtesy Thinfilm.)
posted by Dick Selwood
A couple of months ago, I wrote about ISO 26262 and the changes that this was forcing on the chip development process. (Spaghetti versus ISO 26262 http://www.eejournal.com/archives/articles/20141125-iso26262).
Many of the chips used in vehicles use ARM processor cores, particularly the Cortex-R5, and today ARM has announced that it is making available a safety document set that provide developers with the information needed to demonstrate that their products are suitable for use in systems that meet the highest level (ASIL-D) of safety.
To do this, ARM went back over the entire development process, from initial specification through to final verification. This has been time consuming but as well as providing the material for the Cortex-R5, it confirmed that the development process was robust. It also means that the procedures are in place to produce the safety document sets as part of the normal development process for future cores.
The documentation can also be used for the core safety standard, IEC 61508 and other industry specific standards, such as IEC 62304 for medical products and DO-178 for defence.
As well as hardware, ARM is also supporting software. The ARM compiler is now certified by TUV-SUD as being appropriate for developing software for systems up to ISO 26262 ASIL-D and IEC 61508 SIL-3. Also within the R5, and other cores are functions like memory protection designed for safer software.
During the briefing Chris Turner of ARM came up with something that I hadn't thought of. One of the consequences of ISO 26262 is that there is now a common process and language that runs through the automotive industry, from the manufacturers like Audi and Mercedes, down to the lowest level of suppliers – something that has never existed. This can only be a good thing.
posted by Dick Selwood
At the Future Horizons Semiconductor Industry Forecast on January 20th Malcolm Penn was in one of his classic ebullient modes. His message was "It is time to prepare for one of the strongest (and longest) upswings in chip industry history."
Firstly, the context: this time last year Penn predicted semi sales would grow by 8%, with the most pessimistic case being only 4% and the most optimistic 14%. In fact, with December still to be finalised, it looks like 9.9% for the year.
For 2015 his target is 8.5% growth, with sales of $364.183 billion ($1 billion day). However if we do see a recovery things will be very different. Penn's analysis of the last three upturns showed that the first four quarters of an upturn typically experience growth of around 30%.
However there is not only no spare manufacturing capacity at advanced nodes, TSMC sold out last year, but there is little capacity being built. And even if you start building today, it will be a least a year before you have chips. There is also virtually no inventory being held, all the way from wafers to finished goods. This means that growth will come from price increases as customers compete for the limited supply of chips. The owners of the advanced fabs, mainly TSMC, will be doing well financially.
So while I have been bewailing the lack of excitement in 2014, in one area, if Penn is right, 2015 could become very exciting.