Up until about 18 months ago, things looked very bleak for AMD as the company got absolutely crushed in the processor market by Intel. Everyone knew the Zen architecture was just around the corner, but nobody had an idea whether it would be enough of a revolution to turn the tide. Fast forward to April 2018 and we can see that AMD is once again very competitive in the CPU market. This isn't exclusively because of AMD's execution, it's also because Intel stood still for far too long.
In recent years, Intel abandoned its tick-tock model and the current Coffee Lake generation is the fourth microarchitecture on the 14nm node due to issues with the 10nm process. Almost four months ago, Intel said it shipped 10nm Cannon Lake processors but so far nobody has seen these in the wild. With yesterday's earnings release, the chip giant dashed any hope of 10nm coming anytime soon as it delayed mass production to an undisclosed timeframe in 2019. I presume the lack of details suggest it's no longer on the roadmap for the first half of 2019.
The failure of Intel's manufacturing arm and the impact it had on the delay of future architectures means the company is drifting to a position it hasn't been in since the Pentium 4 days. It's not that Coffee Lake is a bad architecture, there just hasn't been enough progress to maintain a lead over AMD.
The lack of a viable 10nm node means Intel will introduce even more 14nm products later this year. Granted, the company has made big improvements to its 14nm process, but the whole roadmap is still messed up. During the Q1 conference call, the chip giant confirmed 14nm Whiskey Lake for consumers and 14nm Cascade Lake for datacenters. Whiskey Lake and Cascade Lake are expected later this year.
As expected, Intel finally faced some tougher questions from analysts about its 10nm fiasco. First up, in response to a question from Ross C. Seymore from Deutsche Bank Securities, Intel CEO Brian Krzanich replied that Intel is currently shipping 10nm products and that these are the "densest, highest performing products out there." He did mention there are still yield issues with 10nm:
We're slowing the ramp down as we go and fix these yields, and we're able to do that. A), we understand the yield issues. They're really tied to this being the last technology tied to not having EUV and the amount of multi-patterning and the effects of that on defects. But also, the real strength of 14-nanometer, I mentioned in my prepared remarks that we've done 70% improvements in the performance of that technology over its current lifetime. And we believe it continues to have legs, that we can continue to make improvements, both within that process technology and architecturally. That's really giving us the breathing room to go and make these yield improvements.
No details were offered about which 10nm products he's talking about, and when or if we'll ever see these in the retail channel.
UBS Securities analyst Timothy Arcuri went a step further and inquired why Intel is still bothering with 10nm. Wouldn't it make sense to skip the node entirely and jump straight to 7nm? Intel CEO Brian Krzanich replied he doesn't think skipping to 7nm is a good idea:
Okay, so let me try and answer your question. No, there's nothing wrong with the design libraries or anything like that. The proof of that is that we're shipping product. So if there were basic functionality issues like that, you wouldn't be able to produce and ship the product. Again, as I said, this is all around how many layers are on multi-patterning and the end of life of the immersion for the critical layers.
The second part of your question was would it benefit to just skip to 7 nanometers, and would that have an effect on the capital or the gross margins? The simple answer is no. I don't think that's a good idea. The best answer is there's a lot of learning that will happen that we can carry forward into 7 nanometers just like we carried from 14-nanometer to 10-nanometer.