Chipmaking Is Being Redesigned
economist.com
economist.com
It's refreshing to see this mentioned. I'm no semiconductor expert, but it seems weird to me that although node size is measured in a physical unit, nanometers, it does not correspond to any real measurement that exists [1]. Each transistor in a 5nm chip is actually between 28 and 36 nanometers in width. It's called 5nm because of a theoretical calculation based on transistor density [2].
If Tesla advertised a new "200kWh" battery, I would be very disappointed if it turned out that the battery only held 95kWh, but the marketing department had decided that improvements in the charging network have made it "like 200kWh" compared to earlier models.
[1] https://en.wikipedia.org/wiki/5_nm_process
[2] https://en.wikichip.org/wiki/technology_node#Meaning_lost
For comparison, sheet goods aren't priced this way. A 3/4" sheet of plywood really is 3/4", because the plywood was manufactured that way and as a manufactured product there is no seasoning.
You're not being ripped off because a 2x4 is actually smaller. A sawmill had a tree, and they cut off a hunk of that tree, and they charge by how much wood they cut off. The dimensions of that piece of wood changed after they cut it (which in and of itself is a service they provide - it would be a hassle if you had to buy green lumber and dry it yourself).
3/4 plywood will measure less than 3/4.
It's interesting that lumber kind of looks like shrinkflation (what the comment I was originally responding to suggested), but it's more that technical improvements have allowed producing the same finished product with less input material. But we still label the stuff by the amount of input material it used to take for historical reasons, which at a glance looks like shrinkflation.
And what's is worse is how hard it is to know the true size of everything you find in stores in the USA. The same applies to pipes of all types. But even things like a drawer slide is often not the advertised length. Saw blades thinkness is converted from 2.5mm to whatever the closest inch the salesperson felt that day.
From having spent quite a lot of time in big box stores in France and the USA. I can tell you that in the USA 3/4" (19mm) can be anything from 16mm to 24mm.
In short almost everything is built in metric and the USA likes to pretend it's not and gets confused by rounding up randomly.
Perhaps it would be better to go with "6502s per square millimeter" or something like that that would probably be a closer approximation of real-world density, since it accounts for wire routing and so forth.
SRAM density is sometimes used I think.
(Apollo Guidance Computer)
Improved semiconductor device density directly translates into benefits for more advanced computing systems— the primary driver for progress in semiconductor technology. Thus, we propose the use of the following three-part number as a metric to gauge advancement of future semiconductor technologies: [DL, DM , DC ], where DL is the density of logic transistors (in #/mm2), DM is the bit density of main memory (currently the off-chip DRAM density, in #/mm2), and DC is the density of connections between the main memory and logic (in #/mm2). As an example, today’s leading edge technologies that are published in the literature [15]–[17] can be characterized by [38M, 383M, 12K]. As another example, 3-D stacking of multiple logic and memory dies can increase DL, DM, and DC .
Not as cool as megaFonzies.
(still bitter about the whole GB vs GiB think caused by marketing people)
Deciding that in your field you're going to redefine standard units and their abreviations is a bit lazy.
Whether the measurement method is appropriate to your use of an amplifier is of course anybody's guess.
I agree the term "RMS" is kinda miss-appropriated, but it's short and tends to understood as something reasonably close to what it's actually about.
In fact, the draw varies enough during the cycle that it’s a well known method in industrial ecology to put a condenser on a 60Hz motors when the supply is 50Hz. So there must be a visible cycle.
edit: The product blurb says it’s an induction motor. The nameplate says 1000W. It would be fairly straightforward to estimate the efficiency by blending a known amount of water and measuring the temperature change over some period of time.
Accurate RMS can easily be calculated and advertised. Check out Steve Meade and his tools/videos if you're unaware.
Most prominent brands just choose to market absolute bullshit numbers - people who know what they're actually after know how to find the brands who publish real numbers.
In fact, the node number has never referred to transistor dimension, but the length of the gate electrode. A transistor would typically be about 4 times larger than that.
The reason to refer to the length of the gate electrode is because that was always the smallest feature printed on the chip. And that is what defined the required resolution of the lithography process.
So, a lithography process with a resolution of 32nm would be able to print chips with 32nm gate lengths.
Much has changed, and now transistors are no longer planar, they are 3-dimensional fin structures (FinFETs). The gate length scaling has slowed down, and now is not the smallest feature on the chip anymore.
The smallest feature in a modern FinFET process is actually the with of those 'fi s' that form the transistor channels.
And as if by magic, they correspond pretty well to the node names, i.e. a 5nm process will have a fin width of about 5nm.
You will likely not find that mentioned anywhere, it's just a fun fact that I noticed as a VLSI technologist.
I also just checked a handful of processes I have access to, and at least in these cases, the fin width doesn't scale with the technology name (e.g. 5nm and 7nm has 8nm fin width drawn in layout, and model indicates 27nm "width" per fin). This isn't much of a surprise to me, because foundries are just decreasing the number with newer generations of the tech, even if the lithography (the method by which features are etched into the silicon) hasn't significantly changed. For example, 3 generations of a single process were called 7***, then 5***, then 4***.
That's your bias. To a process engineer the parameter that matters is what wavelength you are blasting the mask with and what size a feature it can create.
These processes are not exactly standardised either it would be a major undertaking to take an architecture from TSMC 7nm to 5nm let alone moving from TSMC to Samsung, Global Foundaries or if at all possible Intel.
Not unlike how there are pipeline phases that are just a little longer than they theoretically could be (because all of their neighbors have to line up with each other). If you can fix the bottleneck then everything else can shine, causing a magical, outsized improvement.
Jesus, this is the height of arrogance...
Wendell Wilson isn't qualified to talk about this?
Linus Sebastian isn't qualified to talk about this?
Dr. Ian Cutress isn't qualified to talk about this?
I would listen to all three of those before I listen to anything you say.
So, TSMC is not really that far ahead after all? 28 nm is like a decade old now?
This sounds like women’s vanity sizing. Size 6 is now Size 0.
To make further progress designs had to be changed, and things have gotten fuzzier. Processes from different foundries have different densities. Intel's process for a given node is denser than that of the big foundries like TSMC.
See https://en.wikichip.org/wiki/technology_node for more on this.
1500W PSU!!! But you'll never know how we came up with that figure!
Most established, capital-intensive markets are winner-take-all, where a few big players control the bulk of the market. And they usually remain at the top of the market until some fundamental paradigm shift occurs that causes the market to shrink.
TSMC will be top dog for a while, until something happens that causes the market for custom made silicon to collapse. Just like Intel was top dog until something came along to replace the market for x86. Or like Microsoft was King Dingaling until people moved to mobile computing.
The boring chips made for use in American military equipment are largely produced domestically.
p.s. and thus we have doxxed ourselves just a little bit: age group :-)
Sounds like something Apple would do.
When I joined IBM in 1991, it was still a very inwardly focused company. Although it was clear that things were going to be changing rapidly in the industry, it was hard to get employees to understand how precarious things were - until the big layoffs of the mid 90s.
Besides just diversifying global chip production, Samsung has the (geopolitical) benefit of being in a country with large US bases and plenty of internal demand for their chips (if the Exynos designers can produce something exceptional).
Contrary to many popular technological opinions, Moore's Law will be dead for a while. We're nearly at our physical limits. When this is ceiling is hit, we'll start seeing numerous more competition enter the fray. This is because the research part will be too large for one company to innovate itself out of (TSMC will be the first to hit the ceiling and the competition will reach the same ceiling in a short time. Together, they may advance our scientific understanding but I think governments would need to get aggressively involved). This is afundamental scientific limitation. There is no emerging research that will overcome this problem (zero theoretical research that can be used for engineering the manufacturing needed for the next advancements).
3D transistors are next but that's not growth at an exponential curve. We're looking at a linear growth future for a while.
P.S. I'm 33 and I expect within the next 6-15 years that we'll hit this limit and be at this limit for most of my working career unless we start seeing some seriously massive investment to unearth more advanced physics that we can implement at the pico-level. If you disagree, I'd love to hear what you think that will allow Moore to continue because I've read numerous studies and essentially all have zero tangible implementations.
For example, there's room for a clean, native cross-platform application toolkit that does everything that electron does. But we don't have that. Instead application developers use electron (trading my CPU & RAM for their time). And people get angry at Apple for "only" shipping 16 gigabytes of RAM in their computers rather than angry at lazy app developers. And we don't invest in the tooling we'd need to actually fix the problem. (In this case, Electron but small and native).
Another example - as an industry we know how to make fast compilers (eg Go, Jai, V8, LuaJIT, etc). But instead most new languages (rust, pony, zig, etc) are built on top of LLVM. LLVM used to be super fast - but now even in debug mode its sluggish.
There's usually nothing wrong with old computers & phones, but we throw them out anyway because software developers buy faster computers then take shortcuts.
All these fancy new hardware technologies are not aimed at particle physics, HFT, or fringe high performance shops. They are aimed at consumer grade shitty software and a whole industry who is driven by deadlines and non-technical managers.
The change in the software industry has to come from within.
Yes. I acknowledge the nay-sayers and the history of them on this topic
But to your point of ML being a solution. Honestly, I have a difficult time seeing how ML will give the manufacturing engineers enough tangible/actionable steps to take to make this happen. We need a new physical framework to be able implement an engineered solution. But maybe it's like eating an elephant (one bite at a time).
I will concede this though. IF it were to happen, ML will most likely bring it about. I appreciate you optimism grounded in the technology.
You are talking about fundamental physical limits? Um... No, [1]
"At about 26-28.5 minutes of the Jim Keller:
Jim notes that transistors could reach 10 atoms by 10 atoms by 10 atoms while avoiding quantum effects. People are also working on harnessing quantum effects."
We are so far away from fundamental limits. That even if we assume we could somehow double transistor density every two years, which we dont anymore, there are still at least another 15-20 years before we are close to that limit. And that is not accounting any 3D Transistors.
Realistically we already have Roadmap up to 2030 from TSMC. The only limit we will hit is that the node is too expensive and market could no longer afford the premium. Which You could expect to happen within next 10-15 years.
I watched a bunch of Jim Keller's interviews and I just fundamentally disagree with ~'we'll innovate our way out of it'.
https://www.auto123.com/en/news/the-devastating-effects-of-a...
>"While car enthusiasts the world over are worried about assembly plant closures following the earthquake that ravaged Japan, many are still unaware of the significant role played by companies working at the start of the colossal logistical chain that results in the production of a vehicle.
Did you know that a single vehicle uses from 30 to over 100 chips
to control things like the parking brake, stereo, power steering and safety systems such as stability control? Development of these components is extremely complex, and only a handful of companies are able to meet the demands of the world’s automotive giants."
PDS: The world's automobile manufacturers, that is,
the world's carmakers --
would, or should have, a collective interest in IC/Chip fabrication -- especially in light of the most recent shortage...
trying to sell extended warranty the dealer made sure to repeat that again and again ( i still didn't buy and as of now, 5.5 years later, no chip nor anything else has failed so far :)
That being said, I opted out of it with my last car.
In another car (same mfr) I looked into getting one when the regular warranty was about to run out. Sales guy looked and said: main thing that happens to these older cars is X and Y and you already replaced X so it's not worth it. I always wondered if what he really meant was "main thing that happens is something expensive and I'd rather get paid retail to fix that". Either way I skipped it and never had a problem.
I do consider it insurance.
Most of this stuff is extremely reliable!
Automotive grade chips are probably second only to aerospace parts.
Also, these execs are not dumb, if they could just 'adjust their price and sell 100K more cars' ... well, they've thought of that.
Probably what needs to happen is Merkel, Bojo, EU leadership (and same in other countries) might need to pipe in with something to facilitate the economy, but that's also fraught with risks, it's not like Norway Statoil whereby they just have to 'get the Oil from under the sea and it's bank'.
Manufacturing is a core competency of any car maker (second to supply chain). I wouldn't be surprised to see a fab partner ship between automakers in the future just to ensure they can get the chips they need. This will be at least as much about ensuring old chips don't go obsolete as about supply.
>Mr Duesmann described the problems as “a crisis upon a crisis”. Demand for cars slumped for much of last year because of the coronavirus pandemic, prompting auto suppliers to cut their orders for the computer chips that manage everything from a car’s brakes and steering to its electric windows and distance sensors.
>But demand for cars jumped unexpectedly in the final three months of 2020, as buyers became more optimistic. Audi had its best quarter ever, largely because of a rebound in China.
[0] https://on.ft.com/39V7w9h (paywalled after 3 free accesses)
>Mr Duesmann described the problems as “a crisis upon a crisis”. Demand for cars slumped for much of last year because of the coronavirus pandemic, prompting auto suppliers to cut their orders for the computer chips that manage everything from a car’s brakes and steering to its electric windows and distance sensors.
>But demand for cars jumped unexpectedly in the final three months of 2020, as buyers became more optimistic. Audi had its best quarter ever, largely because of a rebound in China.
You say that development of the chips in cars is extremely complex, but I was under the impression car chips are generally outdated, semi-rugged processors that might have some additional safety features like lockstep cores. The control systems algorithms for the functions you describe aren't particularly complicated as far as I know.
Auto manufacturers are infamous for cost cutting. I assume selling them chips is a low margin high volume business, in which case I would be glad to replace their orders with higher margin ones at the earliest opportunity.
[0] https://on.ft.com/39V7w9h (paywalled after 3 free accesses)
Most of the Chips in Cars are on 28nm if not older. And 28nm Fab Demand has been outstripping supply for quite a while. So 28nm products are getting much more expensive and some customers simply refuse to accept the new price. And when they do, Vendors said it is too late, orders are already taken.
And they just love to blame it on TSMC.
This hasn't been true for what, 15-20 years? Longer? If you had a digital design and enough cash you could have gone to one of many vendors, who in turn worked with an external fab such as TSMC (or in other cases, the fab division of the same company) to create your design. When working with a vendor like Broadcom, Agere, Toshiba, IBM, Intel, Marvel you can be entirely isolated from the physical aspects of making chips, if that's what you want. What has happened over the last few years is massive consolidation of these vendors so now the options are far more limited. It's basically just Broadcom or Marvel at the cutting edge. I don't think this goes anywhere to explain why more companies are designing their own chips but it's a more accurate description of reality.
But I suspect that it is probably "no". Because it looks like ST was planning to pay ~$1.8B per 300nm fab in 2017.
https://www.electronicsweekly.com/news/business/st-build-two...
I assume that "mm" is a typo in the article, but that cost was for a >15-year-old process node. 65nm for $100M might be a stretch, if that is accurate.
The slightly longer answer is "I don't know. But once you have the space and equipment and you're sure you can handle everything safely you still need to staff this and build up some institutional knowledge." If you're thinking this far, you should ask why 65nm? Because it's a process node you know a certain microprocessor was built on? It's just a benchmark that's useful for working out your sense of what it costs to do these from scratch?
The cheapest way to get the equipment would be buying out a fab that's shutting down. When I was in undergrad, I got to play in a 5 micron (5000nm) fab on campus because the previous owner of all the equipment gifted it to the university instead of scrapping it. My gut sense is that there's nothing at 65nm that's unprofitable yet, so you're not getting cheap stuff.
(a 6 micron node was commercialized in 1974, this university lab was opened in 1993. 65nm was commercialized in 2005. While Moore's Law held, more or less, through much of 1974-2005, I think we can say that the difficulties and necessary capital investments increased super-linearly over that time. And 20 year old manufacturing tech seems much more useful now than it did in '93. The STM32 family of microcontrollers, a very strong line, is spread over the 132-40nm range of process nodes.)
The chip industry, and the chip equipment industry are still capacity limited due to demands of remote work and school. That may level off and revert somewhat towards the end of the year. BEV sales may take up the slack, but also may demand a somewhat different chip supply chain and drive demand for different segments in the industry. We may well see the typical crash in equipment sales in 2022 as those two market drivers transpire.
[0] https://www.cnet.com/roadshow/news/tesla-fsd-computer-retrof...
Well, you could have FSD in an ICE car just as easily.
In fact, I'm going to guess an electric vehicle has fewer chips than an ICE car because it doesn't have an engine. Depends how many PMICs you need for the battery cells.
My hunch is now that chip design and chip manufacturing are separate, the cost of chip design is going to be revealed as a lot lower because the opaque accounting of before allowed for stupid inefficiencies.
But in this case, I have no idea whether funding the basic materials engineering, scaling it up, or building the scaled up design, is the hard part.
Just an example, the design has to be altered to maximize yields. You don't know what the yields are like or what to alter in the design until you make a few chips and test them, which means you need to design the test harnesses, tape out a prototype, and get both to where they need to be. And then try again.
If the testing, fab, and design people/materials are in different places you have to move things and people around as a part of the process.
It's just an expensive endeavor that's difficult conceptually and logistically, with a lot of institutional knowledge required.
- Because of the enormous costs of duplicated effort. I'm not sure anti-trust will work.
- Nationalist duplication might, but I am not sure whether it will hinder or hasten WWIII.
- Radical IP disembargo with institutional cross pollination works better for codified then opaque-institutional knowledge.
What is nice about vertical disintegration is is forces the the institutional knowledge to by codified. So maybe things can work in tandem, too.
https://techcrunch.com/2021/01/19/elon-musk-said-it-was-not-...
It's your browser contacting economist.com and fetching the webpage. The javascript app then changes how it is rendered by removing stuff most people don't want to see.
Outline is available as a browser extension, if you don't want to load it by visiting outline.com.
chrome devtools->network tells different story.
This is in the HN guidelines actually ("If the title includes the name of the site, please take it out, because the site name will be displayed after the link.")