Intel foundry customer bails out Another partner silently goes AWOL

[goldstone77]

Intel foundry customer bails out
Another partner silently goes AWOL
Sep 6, 2017 by Charlie Demerjian

It looks like one of Intel’s marquee foundry customers just jumped ship. SemiAccurate has been saying that Intel’s vaunted 10nm process is not commercially viable, now we are seeing the first fallout.


You might recall we have been saying there are two, not one, fundamental problems with Intel’s 10nm process. We told you about the first last December and the second a few weeks later. The problems are not tied together however, fixing one won’t necessarily, or even likely, mean the other is solved, and either one makes for a commercially non-viable process.

You might recall the 10nm process is officially going great according to Intel PR, they talk about Icelake taping out, how they are not behind in Moore’s law because Hyperscaling redefined the numbers, and so on and so forth. One thing they won’t point out is that 10nm products are not out. They were supposed to be out more than two years ago and now, well, 2018. Maybe. And only on the lowest volume parts of the stack, “8th generation” is now overwhelmingly 14nm parts. Does that tell you about how viable Intel believes 10nm is when (if?) it comes out in 1H/2018 as “planned”?


Then there are the foundry customers, lots were announced. Altera would be the big one shipping if Intel didn’t buy them but there are others with silicon on the market. Netronome, Archronix, and Tabula seem to be the entirety of the list but together they don’t add up to more than a blip in Intel’s revenue. Remember Cisco and Panasonic? Cisco signed up in early 2013, Panasonic in mid-2014. Their Intel manufactured devices should be flooding the market now, right? Shall I go on?

Customers love predictability like Intel delivers
So Intel can’t make 10nm chips, has a customer list that can be counted on one hand, and the ones of those with financially significant volumes can be counted on one finger. Progress is being made though, and customers should be happy with the timelines Intel is promising. And delivering. Page 11 notwithstanding. Since the 2013/2014 announcements, several others, a few big names included, have officially signed on. This ‘wave’ was set into motion by ARM announcing they would be working with Intel as a foundry. Big names followed. One of the largest just left.

 

[goldstone77]

Now wouldn't that be something if Intel doesn't have 10nm, and AMD comes up with 7nm. Now, those would be some interesting benchmarks!

 

[Daniel Nenni]

Now wouldn't that be something if Intel doesn't have 10nm, and AMD comes up with 7nm. Now, those would be some interesting benchmarks!

Intel will have 10nm but not for the foundry customers. Apple will have 7nm chips next year but AMD may be a little later. Intel foundry works with ARM on physical IP (SRAM, STDC, and GPIO) not the cores. Unfortunately a big foundry customer uses their own SRAM, STDC, and IO so this is for smaller foundry customers.

Bottom line: When a foundry competes with customers like Intel and Samsung the first choice will be a pure-play foundry like TSMC, GF, or UMS if at all possible.

 

[Portland]

I lived in Washington County. It's a bubble that David Duke would be comfortable in. That's the problem.

 

[IanD]

Just to make it clear, Intel's 10nm process is very similar to TSMC/GF 7nm -- it's not that foundry technology is a generation ahead, just marketing... ;-)

 

[Lodix]

Just to make it clear, Intel's 10nm process is very similar to TSMC/GF 7nm -- it's not that foundry technology is a generation ahead, just marketing... ;-)

we are aware of that here. The problem is that Intel claims to still have a 3 year lead over the rest of the foundries... which is hilarious because current 10nm from Samsung and TSMC is more dense than Intels 14nm and TSMC probably will start mass production of 7nm before Intel has decent yields of their 10nm. And it doesn't stop there, 7nm+ with EUV for 2019 with another shrink.

 

[Daniel Nenni]

Just to make it clear, Intel's 10nm process is very similar to TSMC/GF 7nm -- it's not that foundry technology is a generation ahead, just marketing... ;-)

Yes, Scott summed it up nicely:

https://www.semiwiki.com/forum/content/6895-standard-node-trend.html

Conclusion
By either the old or the new standard node values Intel has lost their multiyear density lead over the foundries. Based on the new more accurate standard node value the average node value for Intel's 10nm and the foundry 7nm processes is 8.05nm and all four companies (GF and Samsung) are within a 0.5nm standard node value of each other.

Intel and TSMC should be about even in 2018 with 10nm and 7nm respectively. I do not expect Intel to have 7nm until > 2020 so TSMC may pass them with 7nm EUV in 2019 and 5nm in 2020.

This Wednesday is the TSMC OIP Forum so we should be getting a quick update.

 

[count]

Meanwhile TSMC announced a 7nm tape out today with Xilinx and Cadence. Intel is starting to look like a short candidate. Without significant volumes from foundry customers, the economics of moving forward on leading edge processes will become increasingly challenging. This, in my opinion, is a tipping point - one that Intel will not be able to recover from.