If anyone's interested in this general subject matter, a while back I did some academic research on highly scalable locks where we came up with some very high performance reader-writer locks:
I love this. Many a moon ago, I worked on a system called Aikido at MIT, which combined a special built hypervisor with a binary rewriting system (DynamoRio) to enable efficient time travel debugging and race detection of parallel applications.
If anyone's interested, here's a publication that talks about it in more detail:
The use of performance counters here also reminds me of another project I worked on called Kendo, which was a posix thread like replacement that used performance counters to enforce a deterministic interleaving of synchronization operations (mutexes, etc). The system could guarantee determinism for programs that didn't have race conditions. Back then, I found that counting instructions wasn't deterministic on the processors of the time, but counting store operations was. If anyone's interested in that work, here's the publication:
Not right now. Sounds like they are working on lenses that could one day work with colored light for cameras. Maybe after that, they could be used for specials?
This article describes how nano metalenses work and what they can be used for.
Spoiler alert: they can't be used to replace your smartphone's camera lenses yet, but can be used for IR distance sensors used on drones and soon, polarization sensors that will be able to tell materials apart and even detect cracks in concrete.
AI-equipped cars with a real sense of feeling in their tires. It’ll be like the sensation of touch for them. Everything they run over, they will feel. Make them learn that animals feel gross to touch. Make them feel remorse for running over a human. Boom! Self-driving safety solved!
I feel like this would be much easier to accomplish with acoustic sensors and wheel/suspension telemetry, given that an attentive human driver can sort of "feel" the road surface already.
It's interesting to see how polarizing this topic is. I'm curious if people are just reacting according to the "party line" so to speak, ie based on whether they are iPhone or Android users.
We can do a poll to confirm this without revealing your phone preference.
Just respond with "Affirmative" if you are an iPhone user and dont like this OR you're an Android users and like this. Likewise, respond with "Negative" if you're in the opposite camp.
I think you are mixing up what you are calling “the party line” with a mixture of - people cheering for the world they want (EU yet again forces a US company to change behaviour), and people who view the chances of this (at best interpretation legally grey) gamble working without that as extremely unlikely.
It's not really that simple. I'm an iPhone use, who thinks Android is kinda pointless, but I don't understand why Apple hasn't opened up the iMessage protocols years ago, or at the very least made their own Android application
"This means processing speeds in devices based on them could reach femtoseconds, a million times as fast as the speeds achievable with current gigahertz electronics"
> “Now that we know what structural and electronic properties are needed...there is good likelihood that we will find earth-abundant alternatives to this rhenium-based material.”
Unfortunately the rarest of rare materials, but the above quote does point out that it shows us what’s possible even if it isn’t viable yet.
Rhenium is used in industry already in metal aloys. It costs around $10000 per Kg. [1] And the US produced 8 tons of it in 2020 [2]. A modern cpu weighs 60g. If we were to make CPUs out of pure Rhenium we could make 133 000 a year at a cost of just $600 each for raw materials. If these things are really a million times faster than the current best in breed silicon ones they will have no trouble competing commercially. $600/ one million is not very much money at all and 133000 times one million is a LOT of compute. Furthermore, the CPUs will not be pure Rhinium.
Gold was extensively used in chip manufacturing previously, and wasn't cost prohibitive.
The 60g figure is for the packaging, not the chip. The actual chips are tiny and weight much less than that, probably under 1g for most processors.
This new semiconductor isn't pure rhenium, it's a compound, and hence less than 1g would be needed, or about $10 per chip, maximum.
Realistically, this new semiconductor would be deposited as an extremely thin layer on top of something cheaper like silicon or quartz. The material cost per chip would be measured in cents.
The cost of depositing an expensive substance like a rhenium compound on a semiconductor wafer is significantly greater than that of the substance that remains deposited on the wafer.
Depending on the kind of deposition method used, for depositing a certain amount on the wafer, a much greater quantity is used, which ends deposited on the equipment, or as chemical precursors mixed and reacted or unreacted.
Due to the rhenium cost and scarcity, all the rhenium compounds that are not deposited on the wafer must be recycled. That can raise the cost a lot.
Finding a compatible substrate for deposition, with an appropriate crystal structure, can be very difficult.
That’s extremely interesting. So there is a considerable manufacturing process waste factor that reduces yield.
I had been thinking that viability of yield could be an issue too as it would have the same wafer fabrication yield, wafer sort yield, and packaging yield that silicon does. And as a new and profoundly expensive material there is going to be an appreciable learning curve.
Is that what you’re referring to? Or is there more to it even than that?
That's not really how market works though, if you could make processors that were a million times faster than current ones, there would be demand for much more than 133 000 a year and many people would be ready to pay more than 600$ for them. For comparison, I find on a random Google result that 50 tons of silicon are used per year for CPUs "in the US".
The price of rhenium would skyrocket to million dollars per kg, and as long as production would stay the same (assuming it's limited by raw resource availability rather than just extraction methods) it would keep getting higher.
The current relatively low price of rhenium (relative to its rarity) is simply due to low demand.
Sure, assuming (1) we can actually produce a rhenium CPU, and (2) we are unable to increase the production of rhenium then, yes, the price of rhenium will increase.
But it will increase precisely because we have a working rhenium CPU in production in the first place, which is what the article disputes is possible (due to its current rarity).
Furthermore, the price would skyrocket only if people are actually willing to pay a high price for these rhenium CPUs, which again means they’re worth the money.
> The current relatively low price of rhenium (relative to its rarity) is simply due to low demand.
For stuff previously needed in low quantities it's usually the reverse - price goes down as more is needed. Initially prices are high because manufacturing equipment has to be maintained even if idle, wages have to be paid, and because of logisticical overhead for the small quanitites. A second price drop occurs as we get into mass-manufacturing and better processes are found.
Rhenium is a byproduct of mining and refinement, but I suspect it is often not captured because the small quanitites needed don't make it economically interesting - you would invest in infrastructure to extract it and immediately crash the price. That would change if there was a stable demand of higher quantities.
Also depending on how easily you can scale these semiconductors down, you may not need such complex cores. CPUs today are complex to (amongst other things) improve IPC - but if you’re really talking a 10^6 improvement in clock speed, I would guess that even a simple in-order core would perform better, and be much smaller.
DRAM latency will remain a problem, no way you can make CPU a million times faster without bringing in memory on the same chip. And you could probably only have a few kB close enough to the core to keep up in frequency.
Most probably you'd see this tech in highly specialized interfacing circuits, like current GaAs chips.
Fiber optic regeneration could benefit from faster electricals on longer shoots, you can only amplify so many times before the noise is amplified too much to ignore so they use optical to electrical to optical circuits. A much faster electrical path could make for some really cool high speed regeneration.
I can also imagine Juniper using them in an ASIC and charging a ton.
If desirable CPUs are made of Rhenium with an effective production line and that the raw supply is so tight, the price certainly won't stay at $10000/Kg, and probably will align with equivalent computation power/watt of traditional designs. Any investor will see that it will only make miner rich without that much return miles aways, and walk away.
This ignore the law of offer and demand. If the supply doesn't change, but there is a lot more demand because now the semiconductor industry also want some, the price will shot up.
processing speeds in devices based on them could reach femtoseconds
That would also mean switching times that allow rectifying infrared light. I.e. capturing light energy with um-sized antennas instead of bandgap traps.
They also seem to say that it cannot be used directly for improving CPUs as they are designed today:
> “[…] they are not necessarily compatible with current hardware used in the semiconductor industry,” […] the applications for these semiconductors “would likely be different than those for traditional semiconductors.”
https://people.csail.mit.edu/mareko/spaa09-scalablerwlocks.p...
reply