Intel should not rename their processes

Analysis: An unwinnable fight should not be picked

Intel LogoOregonLive got a scoop about Intel potentially renaming their 7nm process to 5nm to match the marketing schemes from the foundries. If Intel does this, SemiAccurate feels they are walking into a trap that will accomplish nothing more then making them look worse.

The idea is simple enough, over the past decade or two, process nodes have become completely disentangled from the bonds of reality when it comes to naming. The old micron/nanometer naming convention was based on the smallest line/feature that a node could draw, more or less. Actually lets run with the disclaimer up front, there is a lot of stuff in this article that we will massively oversimplify and some we won’t cover, the broad ideas are what matter, not the technical details. This story is about marketing games that are more or less entirely divorced from anything reality based after all. We will also point out that none of the players in this story have anything close to clean hands.

So back to the story, until half way through the last decade or so, Intel had two leads in process tech. They were a node ahead on lithography, the part that determines the line width aka nm, and also had better tech to use on that node. Intel introduced copper long before their main competition, was ahead on strained silicon, and ended up way out in front on FinFET/Tri-Gate. This meant that when you compared a foundry process to an Intel process on the same node, Intel’s usually performed better. Intel also tended to be ahead on density for released products, the spec sheets may or may not have reflected this advantage.

A key transition point for the industry was the foundry 20nm node, roughly analogous to Intel’s 22nm node. As an aside, at this node and the next, the numbers started to diverge with some using 20nm and others 22nm, then later 14nm and 16nm, but we will simplify things and call it the same node. This is different from the half-nodes that foundries often do but Intel has not traditionally released. Examples of this are 28nm based on 32nm, GlobalFoudries 12nm used by AMD extensively which is a 14nm variant, and the current 6nm generation which is a 7nm variant. Some nodes and foundries have half-nodes, some do not, some call it out, some do not. In any case we will ignore those for now, just be aware they exist.

Back to 20/22nm, Intel had FinFETs and the foundries were on a planar process which was comparatively awful for performance. This limited it’s reach to lower power, lower speed parts and density critical applications. It was a short lived process for cause and almost no one uses it today. Intel’s 22nm with FinFETs on the other had was quite solid, it worked, was applicable to the highest performance CPUs of the day, and all was well.

When the foundries got their FinFET technology sorted, they quickly replaced the 20nm transistors with FinFETs and saw huge benefits in performance and power. This was the 14/16nm node at Samsung, GlobalFoundries, and TSMC. By then Intel was on 14nm with second generation FinFETs and other tweaks at the time, but the foundries were there too, right?

Here is where marketers start spreading slime over, to this point anyway, a mostly technically based naming scheme. That foundry 14/16nm process did replace the planar transistors with FinFETs but they didn’t change anything else, it was the 20/22nm process that should have been but not a shrink. SemiAccurate went over some of the details here but the take home message is that the ‘shrink’ wasn’t one, it was just a name for changes in the underlying tech.

Foundries were years late and Intel was well into their real 14nm shrink with 10nm on the near *COUGH* horizon so their lead was extending. To counter this real lead, rather than calling their 14/16nm node something sane like 20FF, they named it to look like it was a shrink. This was flat out dishonest and started the ball rolling down the hill. As you are probably aware, like politicians, once this set gets away with a little dishonesty, it quickly metastasizes.

Foundry 10nm was the first real shrink from 20/22 and it by rights should be called 14/16 but again, good luck with backpedaling now. Foundry 10nm was a pretty awful node like 20nm, short lived, low performance, and quickly forgotten once their 7nm, another true node shrink, came out. That said neither of these two nodes were very close to the expected 50% area reduction seen in the past, but they were solid enough to not be outright fictitious. The foundries were making real progress on lithography and technology.

About this time, Intel was doing the opposite with their 10nm, it is technically out but 4+ years late and yields are still not functional. We will sum this one up by saying that Intel’s spin went well off the rails and was on many occasions something SemiAccurate would consider dishonest but mostly it was simply intentionally misleading. This delay, still ongoing, meant that Intel went from 1-2 nodes ahead on performance and density of delivered products to 1-2 nodes behind.

This disadvantage was coupled with the 20nm->14nm foundry node naming ganes to make Intel seem even further behind. Those who get the tech understand that as things stand, an Intel node is more or less equivalent to the next smaller foundry node, IE Intel 10nm is on par with foundry 7nm for most things, Intel 7nm to foundry 5nm etc. That said Intel is effectively out of the game until 2023 when their 7nm process could be released so now the foundries have only themselves to compete with.

Until the 10nm debacle, Intel was the adult in the room as far as messaging was concerned and kept the squabbling kids more or less in line. With them out of the way, the real marketing games took off, and they could always point to Intel’s Hyperscaling not-quite-a-lie and say, “They do it too”, when called on some egregious behavior.

TSMC has been the unquestioned leader in process tech for the last 5 or so years with Samsung close on their heels for some of that time. GlobalFoundries pulled the plug at the wrong moment and is out of the game on recent nodes. Samsung’s 7nm node was late and effectively never hit the market in a real way. The current industry wide shortage meant that no one could get enough leading edge wafers and as of this writing, 7nm is sold out and TSMC 5nm is worse. Those that need more capacity for high performance products are forced to move to older nodes as a stopgap. Luckily the marketers are there to help.

A good example of this is Nvidia’s current GeForce 3000 line of GPUs. A few of the high end parts are made on TSMC 7nm and several others are made at Samsung. On what node? 8nm. That is basically 7nm so all good, right? As we mentioned SS7 is effectively AWOL, 8nm is a 10nm derivative made to look like it was close to 7nm. Unfortunately the performance characteristics give it away as we went in to a bit here.

If you talk to anyone involved in selling products related to 8nm, they will all strongly intone that it is a 7nm equivalent even though they know better. The marketing is working. Samsung’s 7nm problems have now led to their 5nm becoming a 7nm derivative rather than a real shrink like TSMC’s 5nm, basically a working 7nm rather than any real changes. Samsung are now years behind on process but right there on node names.

This has devolved into a mess with Samsung being the worst offender showing roadmaps with almost every single digit nm possibility used for something even if almost all the derivatives were 7nm variants. And it changes. TSMC is in the lead on technology by a long way and that gap is increasing with every new node. They are, shockingly, keeping the naming conventions pretty honest too even though Intel is still not in the fight.

Speaking of Intel, this story is about them, right? After the last two or so pages of history and simplified backstory, we should probably get back to them. Intel just announced a foundry wing so in a few years they will be right back in the middle of this PR fight. Luckily for us all they now have adult supervision once again so there is hope that they won’t be a third player in the mud slinging game. Time will tell.

As SemiAccurate linked in the first sentence, there are credible rumors that Intel is going to rename their upcoming 7nm process to 5nm in order to align with the current norms of the foundries. This is a monumentally bad idea. No, we will just say it is just flat out stupid and Intel will lose even more credibility without the hope of gain even if they can point to everyone else and say, “We are just doing what they did to avoid consumer confusion”. There are probably even studies to say how this will help their products on the store shelves, but the end result will be the opposite.

Why? You are fighting with a group that doesn’t care about the rules. If you rename your 7 to 5 to line up with their 5, they can just rename their 5 to 3. Why not, it isn’t based on reality any more. Each foundry PDK update can move the names up to what should be a real full node shrink. Do you think Samsung’s marketing team will lose any sleep over calling their 5+nm node 3nm to make it look like they are ahead of Intel? Intel is chasing a moving target and the competition is faster and in some cases far less ethical, they will lose every time.

You could say they will never learn, and we would agree with that sentiment. Back in the days of Ultrabooks, something SemiAccurate was quite enamored with, Intel tried to pick a similar unwinnable fight. Actually at the time we called Ultrabooks, “Shiny things for the stupid” which is a sentiment we still agree with. What did Intel do then? They took a product that regressed on all technical fronts, often significantly, and cost more, and tried to push it on fashion. Sadly this can work in the real world if it is actually based on the premises being spun on but that wasn’t the case here.

Instead Intel was targeting two things, the MacBook Air and anyone else putting silicon in a PC. The latter anti-consumer actions are now being revived with the Evo program but we won’t go over that again. The fashion side was impossible to win and we literally told Intel just that. Their argument was, paraphrased, some consumers want something fashionable that looks like Apple but also really want Windows. Other than Microsoft employees, and only a few of those, this group didn’t seem to exist but that didn’t deter Intel. They pointed out that their OEMs were going to make fashionable things too and that was going to move lots of expensive but underperforming product.

At this juncture I pointed out to the spinner that Intel would lose and lose badly. A poorly engineered, lower performance alternative made by OEMs that are genetically engineered to cut costs was not going to end well. Worse yet Intel was squeezing their margins unmercifully and pitting them against each other to keep the opposition weak so there was little room to make a good product that also looked good even if the will or the hand holding the stick was there. But this wasn’t the real problem.

The real problem was that the company setting the fashion was Apple. They make something that was, and still is, almost instantly ‘the fashion’. Colored iMacs. White MacBooks. Black MacBooks. Metal unibody cases. Shall I go on? Apple does something and the old thing is immediately old, that is the way of fashion. At times Acer, Asus, and the Taiwanese ecosystem make wonderful products and even a good looking one every now and then. All of zero of these have become mainstream fashion. It is hard to think of an Apple product in the past two decades that hasn’t become mainstream fashion, they can and do set the bar when and where they want. Know what AirPods look like? Seen any clones? That.

Why would Intel lose? Their ecosystem was set to come out with a load of technically inferior knock-offs to the MacBook Air. Do you go into the supermarket and think, “that no-name store alternative brand is probably just as good as the slightly pricier name brand I like”? That is the course Intel was setting. If they ever got close to the mark and had something that threatened Apple, Apple could change it on a whim, the PCs would instantly be ‘old’ and the Apple product would be the new hot item. In two years when the PCs caught up, rinse and repeat. There was no way Intel could reach level ground with this folly much less win, it was a lost cause from the outset. And we directly told them as much then. And it was.

Now we aren’t able to tell the people that matter directly that the process renaming is folly and a lost cause from the outset, but we have a sneaking suspicion some of them might read this article. DON’T RENAME 7NM TO 5NM, YOU WILL LOSE AND DAMAGE INTEL’S REPUTATION. Why? The foundries can and will react, will do so faster, and have less regard to ethics and technical ties. If you must rename, get off the treadmill and call it something different like your internal process names. The best case scenario is that Intel will look like it is playing catch-up and will have to fight the stigma of not being able to catch the competition on merit even if they actually do. There is no up side here, don’t go there Intel.S|A

The following two tabs change content below.

Charlie Demerjian

Roving engine of chaos and snide remarks at SemiAccurate
Charlie Demerjian is the founder of Stone Arch Networking Services and SemiAccurate.com. SemiAccurate.com is a technology news site; addressing hardware design, software selection, customization, securing and maintenance, with over one million views per month. He is a technologist and analyst specializing in semiconductors, system and network architecture. As head writer of SemiAccurate.com, he regularly advises writers, analysts, and industry executives on technical matters and long lead industry trends. Charlie is also available through Guidepoint and Mosaic. FullyAccurate