[Editorial] I read a very interesting report by Daniel Nenni over at SemiWiki.com and thought I would share it and add upon it. The report concludes that Intel’s 14nm node has still not entered volume production. Now this report is not a leak, and is rather an exercise in deductive reasoning by the author. Therefore, take the authenticity of this report as you will; I for one, completely agree with the arguments given.
Intel Broadwell 14nm Has Not Entered Volume Production Yet – Why Though?
Lets begin with a brief overview of Intel’s process nomenclature. Intel’s modern process names starts with P12 as a suffix in a 5 digit nomenclature. The P12 is followed by 2 numeric digits that depict the process advancement in an increasing order. Each successive die shrink is exactly 2 numbers apart. For eg Intel’s 45nm Process was called P1266 and was produced in 2007. The 32nm Process was called P1268 and was produced in 2009. Similarly the P1270 Process is the current 22nm implementation being used (produced in 2011). The 14nm Process, as you may have guessed by now is called P1272. And this is the process that has been giving Intel so much trouble. See, Intel has been following Moore’s Law to the letter and introducing a die shrink every two years, that is, uptill P1272. P1272 also known as 14nm 2nd Generation FinFET (Tri Gate) technology was supposed to start volume production by the end of 2013. Provided you don’t live under a rock, you would have noticed that 2014 is approaching its end and there is still no sign of the 14nm products. In fact, we are slated for Broadwell-K arrival in Q2 2015 by last reports. So just what went so wrong?
This is where the report comes in. It begins by raising the same question I have discussed above: there is absolutely no reason for 14nm to have troubles with good yields as much as they are already having. It points out that Intel has already failed its previous forecast (given on the right) and released an updated forecast (on the left) which points out “Broadwell SoC”, something the author finds strange. If you take a look at the forecasts given above you will notice how much of a difference there is between these two. Intel is currently marketing as still on “Tick Tock” by using the launch dates of processors as the base but in my opinion it is the volume production date that counts. Now I can answer the question that the author raised here. Daniel wonders whether Broadwell is an SoC since the slide clearly says so. But no this is not true. Thing is, Intel must market it self on the tick tock schedule so it is employing a gimmick. The Broadwell SoC referred to in the slide is actually the Intel Core M processors, based on 14nm 2nd Gen Tri Gate technology. These are expected to arrive by the end of September 2014. These will also be known as Broadwell-Y and will have a new revision (revision F) debut by year end.
However, this does not mean that the 14nm P1272 process has yields within the expected range. On the contrary this is very strange news. By now Intel should have shifted the process from R&D to Copy Exactly (Intel’s certified method and technology for producing identical dies in all fabs across the world) and subsidiary fabs. The report states that one possible reason is the change in Materials used, which could have led to unforeseen problems with the process. A very valid argument, one that could easily explain away Intel’s delays. There is another complimentary factor to this argument: the closing of Costa Rica and opening of the Vietnamese plant along with the over all material shift. Something, that would result in massive logistics after shocks. Intel is currently a market leader with its processors so it is in no hurry to push the node forward, and this could lead to ulterior motives that would otherwise not exist.
There is also the problem of Intel having an utter mess of concurrent processor architectures at a given point in time. Here is an example from an editorial I previously published: “Now to adding to the confusion is Intel’s market driven nomenclature of the HEDT processors which does not indicate that it its a generation behind. I also have a feeling Intel might mess with the nomenclature of Broadwell-K and Skylake-S some more to make it marketing friendly. So using current nomenclature and probable naming, an example would be the i7-5960X, the i7-6770 and the i7-5770K all existing at the same time in the market. And nomenclature be damned because if you follow the usual rules, you are going to be seriously misled.
Ofcourse this won’t fool any of the tech savvy people out there but the non-tech savvy people will almost certainly opt for the ‘latest gen’ which, you guessed it, is the locked i7-6770. Enthusiasts like me would have to buy the older generation of Broadwell in order to opt for the unlocked multiplier, while the hardcore audience will be rewarded with an even older architecture. Hmm, I see a pattern emerging.” Intel is also going to release Skylake at about the same time as Broadwell-K and that itself raises interesting questions. Skylake isn’t a die shrink but how can Intel so confidently predict its time line? It is possible that I don’t understand their R&D process correctly but this entire process fiasco smells quite fishy. So is it possible that for the final factor we must consider that the delay is actually artificial in nature? Maybe. Honestly, it is becoming increasingly hard to say anything at the moment.