https://www.nature.com/articles/s41467-025-59675-5 is the paper, claiming "~1.6 petahertz speed." That would be 190-nanometer wavelength, which is into the so-called "far ultraviolet" band of germicidal UVC, 6.6eV photon energy, on the edge of vacuum ultraviolet. And they're switching it with light. So, I wonder how long these devices will last if you keep using them?
They say the light pulse is 6.5fs FWHM, so they weren't able to switch it on and off 1.6 quadrillion times per second; it's just that the transition from on (29nA) to off (<1nA) was only 630 attoseconds long, which is what they're describing as "petahertz". But "petahertz" implies a whole cycle time under a femtosecond, a cycle which would involve two transitions, which would presumably be 1.26 femtoseconds at this speed. (If they measured the speed of the off-to-on transition, I missed it skimming the paper.) And the actual light they're making the 6.5fs pulse out of is a "supercontinuum laser beam that spans over 400–1000 nm". That's still blue enough to raise some concerns about device longevity (though maybe graphene will prove tougher than certain other semiconductors which shall not be named here), but not to the same degree as if they were using actual petahertz light.
I think the 2× exaggeration is sort of forgivable, and nothing else seems to be intentionally misleading, but it would still be easy to draw incorrect conclusions from the headline.
I’m also confused about this. In EE it’s normal to use rise times to calculate bandwidth, but unless I’m missing something they didn’t do that correctly either.
It would be such a strange mistake to occur on a paper about a topic of this caliber that I feel like I must be missing something.
I suspect that maybe the rise time was much slower than the fall time, so it was the fall time they were excited about. But yeah, I'd think a 630-attosecond fall time represents 500–800 terahertz of bandwidth, not a petahertz or more.
My read on the social context: they want to make a VUV/EUV solid state laser (for you-know-what). But they can't make any publishable headway in that direction, so they hype their results with the impressive sounding "petahertz" (terahertz desktop lasers at ambient conditions are mostly already in the bag)
So, ask a trusted peer whether this is a something-burger or an attempted academic pivot. Since life is too short to spend on science you can't call your own: look at invites from ASML/TSMC/SMIC. That's what happened with Sn vapor lasers-- famously fickle things. The company that first got it to work properly (Cymer) isn't even on the relevant wikipedia page? And on the ASML they are associated with DUV? Really pushing the good faith here I think. I first paid attention thanks to interest from ASML.
(Disclaimer: don't know any experimentalists working in the exact domain)
(This is not to say that ultrafast optical switching isn't far more exciting than desktop EUV sources-- in fact, I personally applaud the pivot, if indeed there was one )
I believe I recently saw on goog schol other attempts at vuv lasing/light emission using very similar setup of laser pumping in si-gr (Can't remember the exact search terms I used at the time, or how unsuccessful they were)
E: here's one (2019 dissertation; ndyag ns-pulsed laser probe) which openly states the underlying motivations. You can also check the phd thesis when it comes out, usually they won't hold back on their informal motivations at defence-time :)
If I hadn't read your doubts about the stability above, which reminded me of the limited lifetime of sn vapor sources (there due to contamination of optics), I wouldn't have imputed those motivations to TFA
E2: one from 2019 on VdW vuv detectors (ns response time)
The laser they used seems to be entirely VNIR, stopping at 400nm. I don't think you can pump a VUV laser with longer wavelength light without something like a frequency doubling crystal, and I don't think anyone has done that. So I don't think they were trying to generate an EUV/VUV light source. The only thing about this setup that suggests EUV/VUV to me is their petahertz claim, which I think is an error stemming from getting overexcited about their measured fall time. But I'm not that familiar with the area, so I could be mistaken.
This is some great research, the paper is here: https://www.nature.com/articles/s41467-025-59675-5.pdf and there are two things that stand out in it, the first is that they used a "commercial graphene transistor" and the second is that their apparatus didn't need to be super-cooled or under tens of atmospheres of pressure or in vacuum etc. For me, that means that the risks of bringing this into an actual thing are much less than they have been for other technologies (like Josephson-Junctions).
It's also kind of funny that you could mine the shit out of Bitcoin with something like this, which would either pay for itself or crash Bitcoin, hard to predict.
This is an optical transistor, meaning that a current is controlled with an optical pulse. That means that you can't pipe these things into each other, unless you can build an equivalently fast and efficient light->charge transducer (i.e. a photodetector). Moreover, this physically can't be scaled below approximately the wavelength of the laser (meaning at least 10x larger than CMOS transistors).
It might turn out to be great for the applications that they point out in the paper itself, not so much for logic. I would say bitcoin mining discussions are a bit premature, and potentially not relevant.
> That means that you can't pipe these things into each other, unless you can build an equivalently fast and efficient light->charge transducer (i.e. a photodetector).
>(meaning at least 10x larger than CMOS transistors)
at petahertz (10^15) speeds, you could sacrifice a lot of space for larger components, and still come out on top vs gigahertz speeds (10^9) by doing more work but a hell of a lot faster, no?
if you can build a chip that's a million times faster, you can sacrifice 3/4 of that speed to doing more work with fewer components and still be 250,000x faster.
it would be really interesting to see how this played out. the entire way you build circuits changes. e.g. current adder designs use extra transistors to save carry propagation latency, but for optical, that might make the latency worse...
Propagation delay is not the same, electrical signals travel much slower in semiconductors than light in a vacuum. If you could make an entirely optical chip, size would matter a whole lot less because light will travel much faster through whatever that material will be.
Regarding BTC, just because a single transistor could run faster doesn’t mean it’s necessarily a good candidate for BTC mining (eg. Most mass miners use ASICs, not the fastest CPUs because they care about the fastest way to check LOTS of candidate hashes in parallel, not a single one at an extremely high rate). A light based transistor would either require all of the rest of the system run on light or it would have to be mated to silicon/electrical hardware which would slow down the clock rate of the system.
BTC Mining adjusts based on difficulty. IIRC (I haven’t looked into the specifics in a few years) the protocol will adjust so a block should be mined every 10 mins or so. If there is a massive leap in hardware capabilities, the protocol automatically throws A LOT more difficulty at the mining problem.
One way to crash BTC using a massive hardware advance: dominate the mining for ~ 2 weeks, long enough until until the mining difficulty is adjusted upwards and stop mining. Much higher difficulty + only older hardware doing the work = very slowly processed blocks and a growing backlog of transactions. Eventually the protocol would reduce the difficulty to match higher block transaction rates, but the attack could be maintained by thrashing the miner on/off just after the difficulty is adjusted upwards/down.
Mined bitcoins are distinct from the mining process itself: even after all bitcoins have been mined, miners will continue to operate and earn transaction fees for validating and securing the network.
I guess we’ll see… I mean, bitcoin transactions will have to be valuable enough that we’re willing to pay extra (vs credit or debit card transactions) to maintain that network, right?
If there are fewer miners those that are there get a bigger share of the fees. So this self-balances. As long as there is some value in BTC and some transactions occurring there will be a reward for validating transactions.
>> bitcoin transactions will have to be valuable enough that we’re willing to pay extra (vs credit or debit card transactions) to maintain that network, right?
Does handing transactions require similar amounts of power to mining?
Edit: also, if transactions are ultimately handled by just one or two entities, there will be no point to bitcoin any more.
> The real answer hinges on the leverage of mining one hash to certify multiple transactions.
That's exactly how mining works.
A block is mined when all the data for a block (which includes all the transactions of the current block) plus a nonce gets hashed and the resulting hash has a value that satisfies the current mining difficulty level. If hash doesn't satisfy it, you try a new nonce. Mining hardware just tries millions/billions of nonces per second.
It's possible (though extremely unlikely) that you could solve a block in only a single hash.
I think what a lot of people don't understand is that the difficulty scales with the amount of hashing power on the network. If blocks are being solved too quickly, the difficulty rises. If they're too slow, it goes down. Difficulty really just changes the odds that your hash meets the requirement. It doesn't change the actual difficulty of computation, just the odds of success.
I’m not sure what the real question is or what the real answer is.
I’m under the impression that “handling transactions” in bitcoin and mining are the same thing. (Although I don’t work in cryptocurrency so maybe the misunderstanding is on my part…)
- Broadcasting the transaction throughout the network so that all nodes (including miners) are aware of it
- Miners confirming transactions by solving a block
When a miner solves a block, they earn both the block reward (Which will eventually become zero) and the fees paid by the sender.
Theoretically, as time goes on an the value of Bitcoin goes up, the value of the fees will be high enough that it will make up for the lack of the block reward.
So the original question of "Does handing transactions require similar amounts of power to mining?"
The answer is basically yes. Mining is what confirms transactions. A new block is added to the chain, with each block containing several transactions.
Doesn't bitcoin then completely fall over when the block reward is 0? What incentive do former miners have to process transactions?
I've seen discussion that as the block reward falls, transaction fees will go up but this doesn't seem like enough. Hash rate will fall and people will sell up and get out of it. If the hash rate falls enough doesn't that open the network up to fraudulent transactions if a big player (in terms of hash rate) decides to go rogue? The whole 51% takeover thing. Seems like the entire market cap of bitcoin will pop at some point in the future.
First off, it's not like the block reward goes to zero overnight. It steps down over time.
The reward is currently 3.125 btc. It gets cut in half every 210,000 blocks, which is ~4 years. It'll drop to 1.5625 in 2028, 0.78125 in 2032, and so on. It won't hit 0 btc until the year 2140.
Secondly, even with 51% of the hashing power, a rogue mining org can't create fraudulent transactions. A block with invalid transactions will be rejected by the rest of the network.
Instead, it does allow a rogue miner to pick and choose which transactions it confirms in a block. Still bad, but it's not like they can steal bitcoin from people. And if they just sat and mined empty blocks, they wouldn't earn the transaction fees. The only way they could profit is to short the hell out of bitcoin, start mining empty blocks which effectively is a DoS on bitcoin, then cover the short after it plummets in value.
To do such an attack wouldn't just require immense money to buy the mining equipment, but also a ton of money to put up as collateral for the short.
It’s worth mentioning that BTC was not guaranteed to grow by the orders of magnitude that it did and that some people actually had to use it for real world transactions for it to gain some traction. At least one pizza had to be sacrificed for BTC to become what it is today.
I didn't say I didn't make mistakes either :( had ~70 coins when they were $4 each. Made a killing selling off most of them at $12!! (ugh my heart hurts)
Here's what ChatGPT says about pasting comments copied from ChatGPT:
It signals that instead of contributing your own thoughts, you’ve outsourced your opinion to an AI. In serious discussions, it can undermine your credibility, especially if the topic requires critical thinking, personal insight, or expertise. People might dismiss it outright with a “So what? That’s just what an AI generated.”
It can also feel like a cop-out, especially in academic, professional, or nuanced debates where showing your reasoning matters. Relying on ChatGPT as a crutch might make you seem disengaged or unwilling to do the work of forming your own view.
In short: pasting “Here’s what ChatGPT says…” can make your comment look generic, weak, and uninspired—and in some circles, it’s almost a conversation killer.
Yeah I dislike it when people just paste chatgpt outputs. If you use ChatGPT, you should (imo) use it to _help_ you write your own response - it's output is similar to a Google search, and you use that info to edit your own response
I wasn't using it to prove a point, I was shocked at what it said and trying to find out where it is wrong. But yeah, point taken, my comment's formulation was unfortunate and a bit lazy.
IMO it's all unsustainable. Users will not pay enough transaction fees to match the current block reward, and while mining will continue, if more than 50% of mining power becomes unused, a bad actor can buy it up and anyone with more than 50% of mining power has the ability to completely destroy the network. Tail emission is a better model.
Thank you. While admittedly lazy, I wasn't trying to make a point with "here's what ChatGPT says" - I looked up the economics of BTC for the first time and I was shocked that its demise is built into the protocol (I didn't use ChatGPT to prove a point, I was more trying to find out where it is wrong).
Seems that if nothing changes, mining may become uneconomical as soon as 2028. If that's the case I'm kinda' shocked that there's no panic among BTC holders.
> It's also kind of funny that you could mine the shit out of Bitcoin with something like this, which would either pay for itself or crash Bitcoin, hard to predict.
Bitcoin has a built-in mechanism to counteract improvements in hashing speed (either because of hardware getting faster, algorithmic improvements, or more hardware getting devoted to hashing).
See https://en.wikipedia.org/wiki/Bitcoin#Mining: “The difficulty of generating a block is deterministically adjusted based on the mining power on the network by changing the difficulty target, which is recalibrated every 2,016 blocks (approximately two weeks) to maintain an average time of ten minutes between new blocks“
I think there’s more than enough range available here to handle a million-fold increase in hashing power.
2016 blocks is a lot though. that's nearly $700 million in mining fees.
if someone had a monopoly on chips like these, they could dominate the network and freeze out other miners. which would likely tank the network and make those BTC worthless
That is a risk but not a new risk. It existed with the transition to ASIC miners, too, and didn’t happen.
I guess that didn’t happen because making ASICs is relatively easy, but even if it isn’t, what would be in it for a potential monopolist to tank the value of bitcoin in that way? They better take a large but not overly large part of the market and keep mining money for a long time, only speeding up when a competitor steps in.
> what would be in it for a potential monopolist to tank the value of bitcoin in that way?
Shorting BTC. But you'd have to gather a humongous mining farm / pool in secret, wait until right after the recalibration and then turn on everything at once, mining those 2016 faster. People would panic, enabling you to close your short positions.
After those 2016 blocks you can just keep mining for a regular return, although I'd just sell the farm to an organisation interested in doing so.
It's a fun theoretical attack, if capital intensive. And aside from the short positions (which you can cover instead of going naked), your capital isn't at risk because the worst result is that you'll own a crypto mining farm.
> because the worst result is that you'll own a crypto mining farm.
Bitcoin is fairly popular among criminals, so you might also get some people visit you to argue that is best for both of you to give them back the money they lost, using strong arguments such as threats of bodily harm.
So if you had a machine that could solve blocks at a tremendous pace, could you rapidly go through 2016 blocks with minimal transactions processed, then just switch your machine off?
The difficulty will have gone through the roof and transaction processing time by the rest of the network will then slow to a crawl.
Even with difficulty adjustments you could mine a shit ton, you would basically be mining all of the blocks every 10 minutes for yourself. And if the advantage is significant you could do a 51% attack.
Isn't that assuming you could pack enough of these PHz transistors to make an asic capable of calculating SHA-256? That's quite an endeavor if they have just created one.
Has anyone even made a flip-flop or latch with any optical transistor yet?
Since you asked, yes, optical flip-flops have been around for decades.
That said, you don't need flip-flops or latches to calculate SHA-256 for mining Bitcoins. You only need them at the edges of the circuit, to use the results. But you can do that with electronics at the edge, if you want to avoid stateful logic in the all-optical part.
In 630as light travels half a micron. If that's the clock cycle, a chip would need to do some amazing coordination for bits to reach gates at the same time, and there would be many many cycles before a signal reaches the other side of the chip. Bonkers.
> A study published in Nature Communications highlights how the technique could lead to processing speeds in the petahertz range – over 1,000 times faster than modern computer chips.
A 1 petahertz chip would be 200,000 faster than a 5 gigahertz chip. You've skipped past the terahertz range.
Yes, the fact that contrary to what the title claims, at this point there is no transistor working at petaherz frequency at all. All there is, is a promising new technology.
This is a laser controlled device. Even the terminology of "interconnect" is not really applicable. Your best hope is an optical waveguide coupled to the device, definitely not a metal line. It's not even a transistor in the traditional sense really.
This has limited applications. It doesn't have a viable path to being used in a CPU or GPU. So we're not going to see a zillion-fold increase in compute speeds from this. Maybe some physicists find it useful for an experiment, but the average joe won't notice anything different about the world.
OP's link is about a photonic transistor using graphene. Your link is about making interconnects using individual LEDs and fibers in parallel instead of putting multiple wavelengths on one fiber. They are only superficially related.
So rather than electrons flowing through regular transistors you would have photons flowing through phototransistors? Wouldn't one problem be casting light rays that with widths in the nano or picometer range?
It's not clear at all what path the photons are taking. I read it at first as them travelling as a standing wave, blocking the electrons until the transistor "flips".
If the path of the photons is indeed transverse to the flow of charge, millions of transistors could share a single wavefront.
I used to think electrons flow through circuits, then I learned they actually move extremely slowly, so when you flip a switch there is no way an electron made it all the way to the lamp and back. So now I assume the energy of an electron is transferred to the rest through the electromagnetic field in the circuit or something like that? Honestly, I don't have the slightest idea how any of this works and what is true and what is false anymore. It's not really relevant to computers and yet it's all very fascinating.
It makes me raise so many questions. 1 PHz corresponds to a wavelength of 300nm, UV light. How does it make sense? It can't be the transistors we are used to, that's all quantum weirdness at this point. How do you even use them? Things like copper wires feel meaningless at these scales.
They say the light pulse is 6.5fs FWHM, so they weren't able to switch it on and off 1.6 quadrillion times per second; it's just that the transition from on (29nA) to off (<1nA) was only 630 attoseconds long, which is what they're describing as "petahertz". But "petahertz" implies a whole cycle time under a femtosecond, a cycle which would involve two transitions, which would presumably be 1.26 femtoseconds at this speed. (If they measured the speed of the off-to-on transition, I missed it skimming the paper.) And the actual light they're making the 6.5fs pulse out of is a "supercontinuum laser beam that spans over 400–1000 nm". That's still blue enough to raise some concerns about device longevity (though maybe graphene will prove tougher than certain other semiconductors which shall not be named here), but not to the same degree as if they were using actual petahertz light.
I think the 2× exaggeration is sort of forgivable, and nothing else seems to be intentionally misleading, but it would still be easy to draw incorrect conclusions from the headline.
reply