Limits on Technology

The other day someone made a comment in a thread about A.I. that electrical power would be the limiting factor in future advances in technology. The context of their statement was, to paraphrase, that the capabilities of A.I. would ultimately be limited by electrical power capacity.

I started thinking about this statement and then decided to do some research into it. I was more interested in general technological advancement more so than A.I. in particular. This led me down a pretty interesting rabbit hole. Without going into exhaustive detail about how I wound up where the really interesting stuff was, suffice to say I wound up waist deep in the manufacturing processes for semiconductors. Specifically, the manufacture of computer main processors (CPU). With semiconductors, one of the big considerations is 'transistor count' or density of transistors per a given area of a semiconductor. Today, these densities are measured in nanometers, and the current (2024) generation is known as the "3nm Process Node", or '3nm series'. For those interested, this equates to about 224 million transistors per square millimeter (and yes, you read that correctly, 224mT/mm^2).

After another series of even deeper rabbit holes (which I won't bore you with), I wound up in the area of "photolithography", or the actual process of making these semiconductors. And yes, I really am headed somewhere here, so bear with me. Turns out the current photolithography, a process known as "UV lithography" (technically 'Immersion lithography', or argon-fluoride laser (ArF lithography)) reaches it's limits around 3nm. Next generation chips will require a different process called "Extreme UV lithography" or EUV. What's interesting here is the power requirements. (See? I told you we were actually going to a point with all this).

MOFSET scaling, or the size / density of semiconductors, 'shrinks' in increments, and without going into too many details, the most recent increments are 14nm (circa 2014, these were your Intel i7, X-box and Apple Fusion CPUs), 7nm, 5nm, 3nm (today)...and 2nm (future, like 2025/6).

So, cutting to the chase here; in order to get to the next generation processors, manufacturing technology will need to migrate to EUV type processes. And, in order to do this, the power requirements increase from 1.3kW up to 1.3mW, and the production rate drops by more than 50%. So, let's translate a bit here. In one hour, an EUV process can produce (136) chips, BUT it will take over a megawatt to do it. An average power generation plant is about 100 megawatts (which powers several average cities). Just as an example, in 2023 there were 241 million computers sold (just computers). In order to accomplish this, it would require (2) 100 megawatt power plants solely dedicated to just manufacturing the CPU's for this many computers, and nothing else. For the record, that's 10x as much power as is currently required. And all of this is just around next generation PC's as an example (not all the other 'smart' stuff out there like billions of phones, TV's, refrigerators and whatever).

Summary - It's not only A.I. which will be limited by electrical power, all future technological advances will face the same challenges. And, when people start talking about solar, and wind, replacing fossil fuel and nuclear power, well, I think you can plainly see just how far those sources will have to mature in order to meet the power demands of the future. (A very, VERY, long way...as in 100 years or more).

edit - Update - An average size power plant is around 500mW, so might have to scale down a little for the ratio, but you get the idea.

Two words... Nuclear Fusion.

In 2023 U.S. scientists recreated a fusion breakthrough for the 2nd time.

www.reuters.com...

And in 2024 the money starts coming in to fund these power plants.

www.npr.org...

The future is now.

a reply to: TinfoilTophat

That's awesome, but there are are still some challenges. This process requires massive amounts of electrical power for the lasers, and the reaction is not self-sustaining. So, while there is a net energy gain (which is great), the energy gained is proportional to the energy input. And, when you scale this up to grid level power requirements, there simply isn't enough electrical power to sustain the process for consumption.

Edit - Remember, the output has to perform some 'work' in order to generate power (heat water to steam to spin a turbine, to turn a generator, etc.). Inefficiencies of the system aside, there has to be a mechanism to generate the power.

I'm all for further research in this area (have been since 1981), but as we all saw with "cold fusion" back in the day, you have to study the process very carefully to make sure there isn't something confusing your result. I have confidence this is exactly where Lawrence-Livermore is looking as we speak. Let's hope this works out. I know I sure do.

Two more words... Quantum Computer.

Moore's law says "the number of transistors in an integrated circuit (IC) doubles about every two years". That isn't sustainable and that's why quantum technology is so important. I don't know what kind of power it takes to manufacture, or even run, quantum machines but we'll find out soon enough. Once that happens though, AI could become what people fear. A self-learning, independent being.

a reply to: LogicalGraphitti

Two more words...

Logarithmic and utopia.

First, what we see with my OP is that Moore's law is logarithmic, not linear as many initially thought. That's one of the problems with logarithmic functions; they don't always reveal themselves right away. It's easy to get fooled by logarithmic data sets into thinking they might be a linear function (or even exponential)...right up until you start getting data which shows the downturn.

Regarding quantum computing, well, I don't think that will ever scale. Yeah, they can seemingly make some of the concepts look like they might be viable...at the sub-atomic level...but I doubt they'll ever be able to scale these concepts up to physical devices, not which mimic the theoretical models anyway. Thus, the 'utopia'.

a reply to: Flyingclaydisk

Interesting post and well done sticking at the maths there
Seems like an enormous problem on the horizon, even before we get to quantum tunneling

In a recent podcast, Trump promised to unleash a tremendous amount of energy if elected President.

Do we even need 2nm chips, especially, if they require 1,000 times the amount of electricity, if that's true?

a reply to: Flyingclaydisk

i got to thinking about this after reading, first thought about what kind of place might use massive amounts of power.
then i thought about hospitals, of which many have their own power plants.

i found a site that said

Inpatient Health Care Facilities / Hospitals average 247,000 square feet and use 31 kWh per square foot.

now i'm no mathematician so my numbers might be off.
if converted is right, kwh to mwh the average size hospital of 247,000 x .031 = 7,657mwh . that's not taking into account peak operation or slow time such as night.

links for the conversion,

Convert Kilowatt-hour to Megawatt-hour

Energy Consumption in Healthcare Facilities

and your right, right now i don't think green is going to cut it,and has a long way to go before it can replace fossil fuels. but i did find that some hospitals are going green.

Renewable energy sources for hospitals

originally posted by: Flyingclaydisk
a reply to: LogicalGraphitti
Regarding quantum computing, well, I don't think that will ever scale. Yeah, they can seemingly make some of the concepts look like they might be viable...at the sub-atomic level...but I doubt they'll ever be able to scale these concepts up to physical devices, not which mimic the theoretical models anyway. Thus, the 'utopia'.

Quantum has challenges but that's never stopped research and I expect there will be breakthroughs. I'll probably never see quantum computers on the shelf at Best Buy in my lifetime but it will happen at some point.

About the logarithmic aspect, I'll have to give it some thought. Great topic by the way!

a reply to: Flyingclaydisk

easy peasy. Just put up some solar panels and a windmill. Problem solved. Or does that technology not produce enough power for this?

a reply to: network dude

this is a couple of solar farms FPL has in NW Flordia

74.5 Megawatts of clean, American-made energy

Florida Power & Light Company (FPL) is working to help fuel job growth and move Florida toward energy independence. To date, FPL operates dozens of solar energy centers across the Sunshine State, each of which quietly generates clean, American-made energy for Floridians.

Features of a solar energy center:

74.5 MW solar energy center
More than 200,000 photovoltaic (PV) solar panels
340-acre solar site situated within a 366-acre property
Located in north Santa Rosa County in East Milton, off of U.S. Highway 90

74.5 Megawatts of clean, American-made energy

and a couple of old articles,

Gulf Power/FPL Starts Construction on Two More Solar Energy Centers

Photos: FPL dedicates new Solar Energy Center in North Escambia County

solar and wind farms will have to be massive to furnish all power needed in the future with the technology in use now

a reply to: network dude

The question should be, does solar and wind power produce enough power to make the panels and windmills? If you take the mining of materials, transport, land taken from other uses, and disposal into account, it really doesn't.

a reply to: BeyondKnowledge3

Not even close!

In fact, solar and wind can't even pay for itself without including decommissioning and disposal. It can't even pay for its own capital investment. Actually have a funny story about this, but I'll have to do it in another post. Conversation I had with a solar sales guy once (ROI related). He was selling residential stuff, and didn't expect to run into someone who had detailed stats on big commercial installations (like 100-300 acre sites). Pretty funny conversation. (I think he quit his job after that! LOL!)

a reply to: BernnieJGato

Plus, the grid infrastructure isn't set up to handle this. Infrastructure "tapers" by demand, so the infrastructure to carry the power isn't where the power is being generated. Where the power is being generated is the opposite of where the consumers of that power are. This level of infrastructure costs big bucks to build.

Again, probably a topic for a different day, but the whole thing is just completely upside down. I've written about this here numerous times.

I guess I am old. I do not need technology to survive. I grew up on a farm, you do not need technology to raise animals or plant and harvest veggies. It is nice having a rototiller, mine is a nineteen sixties model. It tills well yet. Fertilizer does not need too much newer technology to make. they have been making it for many decades before computer technology came out just like they made cars before technology came out. The old dodge cars with the slant six and little technology got over thirty miles to the gallon, my sixty nine mustang with a six fifty double pumper. headers, and glass packs on the duels got twenty five miles per gallon but the cost of replacing tires was high since there were black strips all over the roads from burning out.

I am not impressed with where all this technology has brought us, I did not mind cranking windows up and down with a hand crank on cars, I rarely use the backup camera on my car to back up.

It seems that people are starting to believe we cannot live without technology...which is a danger to human existaqnce. It is actually fun getting veggies out of the garden, and fun planting them too. I think a tomato plant looks way better than a flower myself with it's little red cherries hanging on it. I would rather walk in the woods near the road to pick berries than to go buy them in a store.

I asked at the store if they could sell anything without power to run their cash registers, they said no, they have to shut down the store if there is no power...no prices are even on most items you buy and a lot of things are dependent on the computers so the store closes....they have back up power for about half an hour at that store. Some do have generators, but it would cost a fortune to get a generator to run that store and they have to compete with chain stores for price that sell way more products.

Technology is putting us at more risk for disasters effecting more people. Last time I asked at the store about this, they told me one register could still work but now, they have to shut down all together. also was at Meijers buying stuff one day with the wife, and the only people they could check out were those who paid cash because there was problems with their internat which was used to process credit cards and bank cards...not even one of those old rolling credit card swiping
machines to use if the lines were down like they used years ago. Technology is putting us at more risk of a disaster.

originally posted by: IndieA
In a recent podcast, Trump promised to unleash a tremendous amount of energy if elected President.

.....

Trump also said Mexico would pay for his wall; that trade wars are good and easy to win; that rising sea levels will create more oceanfront property; that he would end Obamacare and replace it with "something terrific".

Sounds like the batteries in your bull# detector need to be replaced.

originally posted by: rickymouse
I guess I am old. I do not need technology to survive.

I'm old too but I started my IT career in 1980. A lot has changed but I've spent my entire adult life in that business and I'd probably have to go through some withdrawals if it all went away.

... the only people they could check out were those who paid cash ...

Which brings up a different problem, not many clerks can count change anymore!

a reply to: rickymouse

... Technology is putting us at more risk of a disaster.

I completely agree with you (even though I work in the technology side of aviation). The only thing I would append to your statement would be..."Our ever increasing dependence on""...at the front of your statement.

Society's dependence on technology has gone from simple convenience to outright laziness, greed and stupidity.

So, you are absolutely right!

originally posted by: Flyingclaydisk
a reply to: BeyondKnowledge3

Not even close!

In fact, solar and wind can't even pay for itself without including decommissioning and disposal. It can't even pay for its own capital investment. Actually have a funny story about this, but I'll have to do it in another post. Conversation I had with a solar sales guy once (ROI related). He was selling residential stuff, and didn't expect to run into someone who had detailed stats on big commercial installations (like 100-300 acre sites). Pretty funny conversation. (I think he quit his job after that! LOL!)

That is why I consider large solar and wind power more of an ugly lawn orniments. They have their applications like on small lawn orniments, path lights, solar fountains, etc... Trying to get any really useful power from them is futile. Add in the batteries, backup generators etc...

That kind of green energy is anything but.

A minor point of order here; we're kind of deviating off-topic a bit here (I'm am guilty as well).

My main point in the OP was to illustrate that Moore's law may not be as linear as previously thought, and may in fact be logarithmic. In other words, it won't be long before the sheer amount of electrical power required to advance technology will simply not be available.

We've already seen examples of this in real-time. The super datacenter in Utah is one prime example, and I have illustrated another example in the OP. One which sits squarely on our developmental horizon not ten or twenty years from now, but rather as soon as next year in 2025.

In one breath people talk about green solutions and reduction in consumption, but technology is leading us in the exact opposite direction...almost exponentially greater.