News Intel Process Roadmap Through 2025; Renamed Process Nodes, Angstrom Era Begins

[PCWarrior]

I agree that transistor naming already sucks.
But, I very strongly disagree that renaming the process "brings it more into line with what they really are".
That's the big lie here. That's what is upsetting. They are bringing the number further from the truth, and directly telling people that the new number is the "right" one. You say explaining, I say gaslighting.

Taking naming that was already misleading, and throwing it over the cliff of "words no longer have any meaning" is not an improvement.

I'm tired of false advertising. I'm tired of tech companies being able to do and say anything they want, whenever they want.
Computers have been around for over 50 years. It's not acceptable that the courts and consumer advocacy still don't understand them well enough to recognize even the most blatant abuses of customer trust.

Traditionally, a new node would represent a doubling in the transistor density. That would traditionally be achieved by shrinking the half-pitch of the planar transistor to around 70% of the previous node’s half pitch. When the move to FinFETs was made and with Intel being the first to do so, their idea (Intel’s idea) was for the nm scale to continue by way of equivalency in transistor density, i.e. the density that would be equivalent to that achieved by a hypothetical planar transistor whose half-pitch size would be what the process node would be marketed as. It is not the first time such an equivalency would be made in technology history. For example, the concept of the horsepower was invented to compare the output of steam engines with the power of draft horses and despite today not using horses for transportation or work we are still using the horsepower unit.

However, traditionally, new process nodes didn’t only improve transistor density. They also improved performance per watt by around 20%. When foundries moved to FinFETs they decided to no longer use the doubling of transistor density as their node defining metric but instead use the 20% increase in performance per watt. It still is an equivalency of sorts with planar transistors. It is the performance per watt achieved by a hypothetical planar transistor whose half-pitch size would be what the process node would be marketed as. Of course, performance per watt improvements can be achieved by internode improvements that don’t necessarily increase density. Intel was the last one sticking to its guns and insisting to use the doubling of transistor density as the metric to warrant a new node name. They kept denoting intranode improvements that would improve performance per watt by the + sign. A move which instead of being applauded for its honesty was laughed at by both the tech press and AMD fanboys.

Anyway, with all foundries adopting the 20% improvement in performance per watt as their node defining metric for several generations now and with Intel getting into the foundry business themselves, Intel finally decided to embrace the foundries’ redefinition of what warrants a new process name and no longer use transistor density. Intel is the last one to blame for this. If you want to blame someone, blame Samsung - they are the worst offenders.

 

[escksu]

I agree that transistor naming already sucks.
But, I very strongly disagree that renaming the process "brings it more into line with what they really are".
That's the big lie here. That's what is upsetting. They are bringing the number further from the truth, and directly telling people that the new number is the "right" one. You say explaining, I say gaslighting.

Taking naming that was already misleading, and throwing it over the cliff of "words no longer have any meaning" is not an improvement.
There's a difference between Seagate successfully changing the legal definition of Gigabyte to "one billion bytes", Vs if somebody released a 500 Gigabyte flash drive with a disclaimer on their website saying "Gigabyte is a registered trademark, and is in no way an indicator of capacity. Actual drive capacity will vary by random, due to the manufacturing process". Then the hypothetical drive would actually just end up being a 256MB Wish scam with firmware that reports 2TB free, or whatever.
Process naming is probably somewhere in between. Maybe slightly worse than storage companies blaming their missing capacity on "formatting" (which hasn't been technically in decades. That's from decades ago when you could change the track layout of floppy disks, even back then there was still tons of other mismarketing).

Units of measurement have real world meaning. They're regulated. If the CEO of Exon decided tomorrow to redefine a gallon of gas to "100 ounces of air", then they would be held both civally and criminally liable. But IBM can a 1440KB (now KiB) floppy and get away with printing 1.44MB on the box, because no powerpoint presentation in the world is interesting enough to teach a government employee why there's no possible interpretation where that claim was correct.
That one was like printing "40ish ounces, maybe by weight" on a quart of ice cream.

I'm tired of false advertising. I'm tired of tech companies being able to do and say anything they want, whenever they want.
Computers have been around for over 50 years. It's not acceptable that the courts and consumer advocacy still don't understand them well enough to recognize even the most blatent abuses of customer trust.

Just to let you know that there is nothing wrong with seagate's definition (both technically, legally and mathematically). In decimal system, giga means 1 billion or 10^9.

Also, you cannot trademark a commonly used word. The person may trademark byte (before computers exists) but not giga. Because it has been used as a term to represent 1 billion long before he was born.

Btw, speed is measured using decimal, not binary. 1GB/s = 1000MB/s.... Using binary it should be 1024 but nope.

 

[escksu]

Just to add 1 more thing. The original "standard" way of measurement was microns, not nm....nm was used later as its easier to say 40nm than 0.04 microns. 7nm would be 0.007 microns.

 

[InvalidError]

Just to add 1 more thing. The original "standard" way of measurement was microns, not nm....nm was used later as its easier to say 40nm than 0.04 microns. 7nm would be 0.007 microns.

'microns' aren't a measurement method or standard, just an SI measurement scaling prefix with the implied unit of meters in this context.

The "measurement standard" for processes is none, each fab uses a different weighed average of features to come up with its own number.

 

[pocketdrummer]

you are actually very wrong, the whole nanometer marketing has become huge misnomer many years ago.

Just to give you an example actual transistor size on 7nm TMSC node is 22 nm,
actual transistor size on 14+++ Intel node is in reality 24 nm in size.

The article does a good job explaing why.
Renaming the process nodes just brings it more into line with what they really are.

I came here to say this. People need to chill out about this as the whole point is to remove the misconception that nm measurements are equivalent from brand to brand. They are only relevant when you compare the same company's process to the next.