CPU makers are experimenting with alternative substrates to double clock speeds

[Shawn Knight]

Bottom line: Heat is a computer's worst enemy, and the latest batches of cutting-edge CPUs rolling off assembly lines are among the hottest ever produced. It's a trend that's simply not sustainable for much longer, and Silicon Valley is well aware. Fortunately, some of the world's top chipmakers are already experimenting with a variety of materials that could significantly lower operating temperatures.

Diamond Foundry, which makes lab-grown synthetic diamonds in San Francisco, is leading the charge. The outfit has produced hundreds of four-inch-wide synthetic diamond wafers measuring less than three millimeters thick. The idea is to replace some of the inactive silicon in a traditional microchip with a layer of synthetic diamond, which is an excellent conductor of heat.

Martin Roscheisen, Diamond Foundry CEO, told The Wall Street Journal that chips using their synthetic diamond wafer can operate at a minimum of twice their rated clock speed without failing. In the lab, company engineers have even reportedly managed to get one of Nvidia's most powerful chips to run at three times its base clock.

Roscheisen said Diamond Foundry is in talks with leading chipmakers, EV makers, and defense contractors to help improve chips and electronics. Key to further exploring this avenue, Roscheisen added, is the falling cost to manufacture synthetic diamonds.

Diamond Foundry isn't the only alternate chip substrate maker in town. A company named Coherent makes polycrystalline diamond wafers while another known as Element Six offers larger pieces that can be used between chips and heatsinks.

In September, Intel introduced a glass substrate for next-gen packaging that it has been working on for more than a decade. Compared to modern organic substrates, glass has better thermal, physical, and optical properties which translate to an interconnect density increase of up to 10x. Glass also offers 50 percent less pattern distortion with enhanced flatness that improves depth of focus for lithography.

Intel at the time said it hopes to deliver its first complete glass substrate solutions starting in the second half of this decade.

Permalink to story.

https://www.techspot.com/news/100752-cpu-makers-experimenting-alternative-substrates-double-clock-speeds.html

 

[user556]

Not a good idea to conflate die materials (diamond) with interposer/packaging materials (glass).

 

[quadibloc]

Doubling the single-core speed of a microprocessor would be worth a great deal, so even if chips this way had to be sold at premium prices because they cost more money to make, it would still be worth making them. As an example, remember the especially binned chips sold to day traders; but a more dramatic increase would have all kinds of applications.
Supercomputers - HPC - would benefit immensely.
A different substrate could best benefit speed, though, if in addition to better heat conduction, it was a low-k material so as to reduce capacitance delays on the interconnects between transistors.

 

[WhiteLeaff]

How do they waste time on something that is known to be so expensive and unfeasible?

 

[Lew Zealand]

How do they waste time on something that is known to be so expensive and unfeasible?

Because that's how advances are made. If everyone pooh-poohed every expensive experiment then tech would be *decades* behind where it is.

It was reasonable to think it would work and it did. Now they can concentrate on trying other cheaper materials with reasonably similar properties, maybe using artificial diamonds or glass, etc. to make this big performance jump financially feasible.

 

[Mr Majestyk]

And here I though Nvidia's pricing meant they were already diamond coating their gpu's. Hell the 4080 uses unobtanium it seems.

 

[toooooot]

" In the lab, company engineers have even reportedly managed to get one of Nvidia's most powerful chips to run at three times its base clock"
Amazing.
Can't wait to have my core 16700 running at 10GHz.
I think these types of improvements will keep progress moving when they can't make transistors smaller.

 

[wizardB]

How do they waste time on something that is known to be so expensive and unfeasible?

Diamond is one of the very cheapest industrial minerals, it is only carbon after all the most common element on the planet it is mined by the ton and only suckers pay a premium because it's shiny.

 

[kiwigraeme]

How do they waste time on something that is known to be so expensive and unfeasible?

Because it's not your money

Material Chemistry - is still far from the edge in realms of possibilities

Physicists still do not even have even the faction of the calculating power to work forwards or backwards for the synthesis on new stuff- ie predict exact properties etc

Yes you are right in one way want 3 times the power - just increase real estate , higher node , new chip designs etc

I sure costs can come down - drills must chew through diamonds etc

I'm sure there are customers - like customers for crazy expensive solar cells

US defense force for one
Plus we will get new cheaper materials - with slightly less performance
Add in once we hit 2nm - you start to need to build up in layers
Those photos of some rando in a lab coat carrying a huge circular wafer - well those wafers aren't cheap

 

[wiyosaya]

How do they waste time on something that is known to be so expensive and unfeasible?

This is not naturally occurring diamond that is dug from the ground and rare. Its lab-grown diamond which is far cheaper, far more pure, and most importantly, it can be made on demand in the lab via relatively simple and straightforward processes that are common to the lab and modern manufacturing processes.

The answer to why they are investigating this material is because of its thermal conductivity - or its ability to conduct heat from one place to another; in this case, from chip to heatsink meaning the chips can run at much higher powers without melting.

With this, perhaps the days of piddling performance increases for 5x the cost to the consumer will be over and, perhaps, no more generations of GPUs like the GTX 4XXX series.

 

[gamerk2]

How do they waste time on something that is known to be so expensive and unfeasible?

That's how advances are made. *Every* new technology is expensive and unfeasible, until rising production eventually results in prices falling, leading to mass market adoption.

And lets fact it: We're nearing the peak of advances we can make on CPUs using current technology.

 

[gamerk2]

Diamond is one of the very cheapest industrial minerals, it is only carbon after all the most common element on the planet it is mined by the ton and only suckers pay a premium because it's shiny.

That, and one company owns literally *all* the Diamond mines on the planet, and thus controls the price.

From a supply/demand standpoint...Diamonds should cost next to nothing.

 

[Vanderlinde]

Is'nt the traditional silicon already made of partial glass? I mean smash or break a core, it will shatter like, glass.

 

[WhiteLeaff]

Diamond is one of the very cheapest industrial minerals, it is only carbon after all the most common element on the planet it is mined by the ton and only suckers pay a premium because it's shiny.

The material used today to manufacture GPU and CPU chips is the abundant silicon extracted from sand, and even so, the refinement processes, manufacturing complexity, etc. involved end up making the process complex and super expensive.

Lab-created synthetic diamonds are generally more affordable than natural diamonds, but they can still be expensive, especially if they are of high quality. I'm not going to answer everyone, but someone gave a good answer, they insist on unfeasible solutions because it's not their money. Likewise all other ideas that die on paper, after years and even decades at the cost of millions from foolish investors.

 

[EdmondRC]

Well, forget Ti, next generation Nvidia GPUs will have designation Di when made with diamond substrate.

The 5080 Di (pronounced Fifty-Eighty Die), Jensen will wear a bedazzled leather coat to announce the new cards.

Price - If you have to ask you can't afford it.

 

[EdmondRC]

That, and one company owns literally *all* the Diamond mines on the planet, and thus controls the price.

From a supply/demand standpoint...Diamonds should cost next to nothing.

This will always require synthetic diamonds most likely; the process will need to be controlled.

 

[user556]

Is'nt the traditional silicon already made of partial glass? I mean smash or break a core, it will shatter like, glass.

Not glass as a material, no. Glass is amorphous, ie: not crystalline. But the same silicon sand is used for making both.

 

[Fastturtle]

Because that's how advances are made. If everyone pooh-poohed every expensive experiment then tech would be *decades* behind where it is.

It was reasonable to think it would work and it did. Now they can concentrate on trying other cheaper materials with reasonably similar properties, maybe using artificial diamonds or glass, etc. to make this big performance jump financially feasible.

It would be even worse: The best tech would still be tubes as used in UniVac and it would require a priesthood to keep it running. No I'm not willing to go back to the 1950's for our tech but I'd be willing for the environment and quality of food.