In the real world 6 would be more along the lines of "hardly any big custom chips".
The whole article does suffer a bit from multiple personalities. How are ASICs going to be too expensive to produce in the future if hordes of Indian engineers will be making them on $50k fabs in their garages and Moore's law will slow down enough that design complexity will no longer increase exponentially with time? Why are the big iron chips still alive if they're dead since they're 1 Billion transistor+ designs?
Being able to create custom chips in your garage is still many years away from being possible... note that the MIT team working on this is aiming for a $50K price for a garage fab, not a lot of people will be able to afford that (startups, yes).
I think what the author is trying to say is that in the meantime the whole custom chip thing is in trouble.
It's cheap enough that anyone with a conceivable need for a chip fab can easily afford one.
Regardless of what the article says (it'll be just like youtube!), I can't imagine why random individuals would want to create their own chips, even if they could.
item 6: no more cutting edge process custom chips. e.g. 45nm
item 7: everyone can create custom chips with relatively gigantic feature size which won't be competitive with the 45nm guys, but may be just spiffy for other functions.
This really is an amazing combination of repeating facts from people who know what they're talking about, and clueless extrapolation from a reporter that doesn't really understand all of what he just heard.
Yeah, I cringed at that, and made me wonder how valid the rest of the article is.
Still, overall it seemed like an interesting bird's eye view of what people are talking about, as someone who doesn't follow the hardware industry closely.
> Intel recently had a chip design team in India create its Dunnington server chip, which was announced last year. That’s one of the most difficult kinds of chips to design.
Huh? What kind of chip is the most difficult to design?
That's like saying Windows XP is one of the most difficult types of Windows to design. That makes no sense.
{edit} There's also this gem: "It will be have 16 processing cores that can do for tasks, or threads, at a time." {/edit}
Yea, but if you look at the guy's resume of places he's worked you'd think that he'd have picked up some proofreading skills along the way... 'It will be have?'
Not the best written article out there, but there are some valid points. The part about design tools not being up to snuff is quite true. As is the reason: not enough money is being invested in the EDA (Electronic Design Automation) field right now. Most of the VC has been going to the web space for years now and that's sucked a lot of the capital away from very important things like chip development and chip design software.
I would also suggest that the brain drain isn't actually H1B's moving back to their home countries: a huge amount of talent has been sucked into the web space. EDA software is one of the most challenging areas of software development - all kinds of interesting problems to solve in chip layout, power minimization, high level synthesis, etc. Yet hardly anyone graduating with a CS degree knows that it even exists - they graduate expecting to work for a social networking, web company. To be fair, it's not like EDA is a big enough field to absorb that many CS grads in any given year - in fact EDA is rather stagnant right now... but I think that goes back to the lack of capital going into EDA at this point. Given how important the semiconductor industry is to the US it does seem like a major misallocation of resources.
The big problem is that the the chip industry's market is shrinking in the long-term. As hardware has gotten faster ('fast enough'), the need for custom hardware solutions has reduced, since you can more easily and cheaply do things in software. Similarly the need for faster general-purpose hardware has also reduced, at least in the consumer space, hence the constantly dropping ASPs of desktops and laptops. Emerging markets like China help, but only while penetration there is still low.
And since we all pick the fastest path to riches, investors as well as engineers flock to Google/Facebook/etc. Combined with us reaching natural limits of Moore's law (using Si at least), we aren't likely to see nearly the same leaps in the next two decades as we've seen in the past two.
Disclaimer: I was in the chip industry and recently moved into webapps with my startup
Has the total chip market started shrinking yet? I thought growth from phones and other consumer electronics would easily make up for any dip in PC/server chips.
Item 6 (goodbye custom chips) is not a new trend. It's been happening over the last several years. Programmable logic has progressed and encroached on all but the high volume ASICs. If the volume is there, an ASIC makes sense. Otherwise, there's a wonderful world of FPGAs just waiting for you to explore.
Item 8 (no more chip startups) may be technically true, but I think it's just an evolution of the business model. Rather than planning for an ASIC immediately, you would begin by implementing the IP in an FPGA. This allows you to prove the design and the market. If the market is sufficiently large, you may choose to create an ASIC.
Item 10 (brain drain continues) seems obvious. Of course the capability of engineers in India and China would improve and catch up to the U.S. With the improved skill there's also an increase in cost as those with the skills commands better pay. Equilibrium will follow.
20 comments
[ 3.3 ms ] story [ 58.2 ms ] threaditem 7: everyone can create custom chips in their garage.
Color me befuddled...
The whole article does suffer a bit from multiple personalities. How are ASICs going to be too expensive to produce in the future if hordes of Indian engineers will be making them on $50k fabs in their garages and Moore's law will slow down enough that design complexity will no longer increase exponentially with time? Why are the big iron chips still alive if they're dead since they're 1 Billion transistor+ designs?
I think what the author is trying to say is that in the meantime the whole custom chip thing is in trouble.
Regardless of what the article says (it'll be just like youtube!), I can't imagine why random individuals would want to create their own chips, even if they could.
item 7: everyone can create custom chips with relatively gigantic feature size which won't be competitive with the 45nm guys, but may be just spiffy for other functions.
"[The Power 7] could probably download the entire iTunes library to your computer in about a minute or two"
Still, overall it seemed like an interesting bird's eye view of what people are talking about, as someone who doesn't follow the hardware industry closely.
Huh? What kind of chip is the most difficult to design?
That's like saying Windows XP is one of the most difficult types of Windows to design. That makes no sense.
{edit} There's also this gem: "It will be have 16 processing cores that can do for tasks, or threads, at a time." {/edit}
I would also suggest that the brain drain isn't actually H1B's moving back to their home countries: a huge amount of talent has been sucked into the web space. EDA software is one of the most challenging areas of software development - all kinds of interesting problems to solve in chip layout, power minimization, high level synthesis, etc. Yet hardly anyone graduating with a CS degree knows that it even exists - they graduate expecting to work for a social networking, web company. To be fair, it's not like EDA is a big enough field to absorb that many CS grads in any given year - in fact EDA is rather stagnant right now... but I think that goes back to the lack of capital going into EDA at this point. Given how important the semiconductor industry is to the US it does seem like a major misallocation of resources.
And since we all pick the fastest path to riches, investors as well as engineers flock to Google/Facebook/etc. Combined with us reaching natural limits of Moore's law (using Si at least), we aren't likely to see nearly the same leaps in the next two decades as we've seen in the past two.
Disclaimer: I was in the chip industry and recently moved into webapps with my startup
Item 8 (no more chip startups) may be technically true, but I think it's just an evolution of the business model. Rather than planning for an ASIC immediately, you would begin by implementing the IP in an FPGA. This allows you to prove the design and the market. If the market is sufficiently large, you may choose to create an ASIC.
Item 10 (brain drain continues) seems obvious. Of course the capability of engineers in India and China would improve and catch up to the U.S. With the improved skill there's also an increase in cost as those with the skills commands better pay. Equilibrium will follow.