What’s down the road for silicon?(nytimes.com)
nytimes.com
What’s down the road for silicon?
https://www.nytimes.com/2022/05/16/science/electronics-silicon-gallium.html
19 comments
Perhaps someone here can answer this, but why are wide-bandgap materials like SiC and GaN considered so promising for power applications but not more traditional semiconductor applications (CPUs, GPUs, SOCs, etc)? Seems like the ability to perform more efficiently and at higher power efficiency should be useful there too.
Is it just an issue of not being able to make crystals of the same defect-less purity that we've achieved for silicon? And is it a fundamental issue, or one that can be overcome by developing a similar level of expertise?
Is it just an issue of not being able to make crystals of the same defect-less purity that we've achieved for silicon? And is it a fundamental issue, or one that can be overcome by developing a similar level of expertise?
Well it is mostly not worth the money. One of the biggest benefits of wide bandgap materials is the higher breakdown voltage. But that should not be an issue for logic circuits -- one should not ever reach the breakdown voltage.
Another benefit of wide band gap devices is the low on resistance. This would also benefit logic circuits, but the benefit would be much smaller because logic circuits move much less current than power circuits. Switching speed would be a definite benefit for logic, but is it worth the cost.
The biggest driver of profitability for logic circuits nowadays is large wafer sizes and small feature sizes. Wide band gap semiconductors are nowhere near silicon in this regard. In silicon carbide the newest thing is 200 mm wafers, available only in a few fabs, while silicon has been on 300 wafers for many years. As far as feature size, Silicon Carbide is absolutely giant compared to silicon.
It would take hundreds of billions of dollar to make silicon carbide features as small as silicon, and nobody wants to pay it because currently silicon carbide is much more useful in power applications where you want large features.
The issues for gallium nitride are similar, except that gallium nitride may narrow the cost gap with silicon because it is currently made by merely depositing gallium nitride on silicon wafers, so a lot of the production technology is similar to silicon. But then again when one starts pushing the edge of the leading nodes of silicon, I am sure there a lot of differences between gallium nitride and silicon will appear, so using gallium nitride will not be that simple.
So you may see gallium nitride appear on some mixed signal chips or some logic chips that are far in the lagging edge (e.g., MCUs?). But currently there is so much demand for this stuff (both SIC and GAN) for power that using it for logic is way on the backburner for everybody.
Another benefit of wide band gap devices is the low on resistance. This would also benefit logic circuits, but the benefit would be much smaller because logic circuits move much less current than power circuits. Switching speed would be a definite benefit for logic, but is it worth the cost.
The biggest driver of profitability for logic circuits nowadays is large wafer sizes and small feature sizes. Wide band gap semiconductors are nowhere near silicon in this regard. In silicon carbide the newest thing is 200 mm wafers, available only in a few fabs, while silicon has been on 300 wafers for many years. As far as feature size, Silicon Carbide is absolutely giant compared to silicon.
It would take hundreds of billions of dollar to make silicon carbide features as small as silicon, and nobody wants to pay it because currently silicon carbide is much more useful in power applications where you want large features.
The issues for gallium nitride are similar, except that gallium nitride may narrow the cost gap with silicon because it is currently made by merely depositing gallium nitride on silicon wafers, so a lot of the production technology is similar to silicon. But then again when one starts pushing the edge of the leading nodes of silicon, I am sure there a lot of differences between gallium nitride and silicon will appear, so using gallium nitride will not be that simple.
So you may see gallium nitride appear on some mixed signal chips or some logic chips that are far in the lagging edge (e.g., MCUs?). But currently there is so much demand for this stuff (both SIC and GAN) for power that using it for logic is way on the backburner for everybody.
Can’t make fast, complementary (P-channel) devices, thus the static power dissipation of n-channel only logic would be huge.
How about photonic circuits?
Photonics doesn't buy you much.
The primarily limit on speed in large chips like processors is NOT the transistors but the transmission lines connecting the transistors. And those are already operating at 50%-75% of the speed of light. The gain going to photonics doesn't justify the cost at this point.
Another factor: pretty much EVERYTHING in current microelectronics gets dumped in the garbage if you were to 100% switch to photonics and that's a step too far. There is relatively little in electronics that can be leveraged to photonics.
In general you need about 80% overlap with the prior major technology to justify a jump that big. This was true of tubes to transistors, and then transistors to ICs. You'd see the exact same circuit used in the prior technology being repurposed to the new one - almost verbatim. And it's that leverage that justifies the inevitable higher cost of any new technology - at least you can re-use designs, skills and concepts.
The primarily limit on speed in large chips like processors is NOT the transistors but the transmission lines connecting the transistors. And those are already operating at 50%-75% of the speed of light. The gain going to photonics doesn't justify the cost at this point.
Another factor: pretty much EVERYTHING in current microelectronics gets dumped in the garbage if you were to 100% switch to photonics and that's a step too far. There is relatively little in electronics that can be leveraged to photonics.
In general you need about 80% overlap with the prior major technology to justify a jump that big. This was true of tubes to transistors, and then transistors to ICs. You'd see the exact same circuit used in the prior technology being repurposed to the new one - almost verbatim. And it's that leverage that justifies the inevitable higher cost of any new technology - at least you can re-use designs, skills and concepts.
Chip interconnect is much slower than that over long distances. In addition to the LC delay you mention, the RC delay is substantial and increases as the square of distance so it dominates for long wires. The distance-squared effect is mitigated by adding repeaters along the way, but they add their own delay.
Overall wire speeds of 40 ps / mm are typical, about 8% of the speed of light.
Overall wire speeds of 40 ps / mm are typical, about 8% of the speed of light.
They're still at the basic research stage, with no real path to practical applications yet.
Still? It's been thirty years
The latest news was development of an optical transistor, but the output signal is electrical, not optical. That would make chaining them to make a logic gate tricky.
You can't transmit large amounts of power through a photonic circuit, so forget about using them to replace power electronics.
How about for information?
Of course you can. Otherwise we wouldn't use fiber optics everywhere. You only need a teeny tiny amount of power transmitted to carry a distinguishable signal after all.
What an odd article.
It talks about the move from Si switching power supplies to SiC... and now to GaN. This has been going on for years to the point that many new high power wall plugs are moving to GaN. It's true that EVs need power efficient supplies and drive part of the market, but I don't see evidence that they've radically accelerated the shift.
I don't see evidence that SiC or GaN supplies are noticeably more available (other than being higher margin and thus more likely to be prioritized) than Si with shipping issues and the Shanghai shutdown. I also don't see evidence that location of volume (lower margin) manufacturing are likely to be different for these SiC/GaN components until geopolitical events push them to. Certainly, component assembly is unlikely to shift to the US.
It talks about the move from Si switching power supplies to SiC... and now to GaN. This has been going on for years to the point that many new high power wall plugs are moving to GaN. It's true that EVs need power efficient supplies and drive part of the market, but I don't see evidence that they've radically accelerated the shift.
I don't see evidence that SiC or GaN supplies are noticeably more available (other than being higher margin and thus more likely to be prioritized) than Si with shipping issues and the Shanghai shutdown. I also don't see evidence that location of volume (lower margin) manufacturing are likely to be different for these SiC/GaN components until geopolitical events push them to. Certainly, component assembly is unlikely to shift to the US.
It's a bit garbled as you'd expect from a non-specialist publication, but I thought it did a decent job of highlighting recent trends in wide band gap power electronics. I think that battery electric vehicles are the "killer app" bringing silicon carbide to the mainstream. The technology itself has been developing for a long time, certainly, but BEVs combine some key attributes to drive SiC demand:
- Efficiency has a high value, since a dollar's worth of SiC electronics can save more than a dollar's worth of battery capacity to reach the same driving range. Vehicle range is one of the key attributes for prospective BEV buyers.
- The devices need to handle high power, which mostly rules out gallium nitride for vehicle power trains at present.
- The market is large and growing rapidly, thereby justifying forward looking investments in new manufacturing capacity (like the New York fab mentioned in the article).
- Efficiency has a high value, since a dollar's worth of SiC electronics can save more than a dollar's worth of battery capacity to reach the same driving range. Vehicle range is one of the key attributes for prospective BEV buyers.
- The devices need to handle high power, which mostly rules out gallium nitride for vehicle power trains at present.
- The market is large and growing rapidly, thereby justifying forward looking investments in new manufacturing capacity (like the New York fab mentioned in the article).
I'll agree EVs are scaling the business since they use very large areas (roughly proportional to peak power) and they certainly have high marginal utility due to battery cost. In particular, just 1million EVs with 300kW supplies are 300GW of power supply, while 200Million mobile chargers at 80W are only 16GW of power supply area.
Still, the size of the market is only ~$1B growing to $3B over the next 5 years.
It's also worth noting that GaN can also be grown on SiC wafers so there is utility even when the industry shifts to even higher performance. I didn't realize that 5G base stations were also a significant consumer of GaN.
https://www.yahoo.com/now/sic-wafer-market-growth-trends-100...
Still, the size of the market is only ~$1B growing to $3B over the next 5 years.
It's also worth noting that GaN can also be grown on SiC wafers so there is utility even when the industry shifts to even higher performance. I didn't realize that 5G base stations were also a significant consumer of GaN.
https://www.yahoo.com/now/sic-wafer-market-growth-trends-100...
If efficiency had high value, cars would look very different. We’ll end up repeating the same ”I’ve got a V8 F-350 for my office commute” thing with electric vehicles
Bosch opened a new fab recently.
https://www.golem.de/news/halbleiterwerk-bosch-beginnt-bau-n...
Couple of things that I do not agree with: I am not sure whether the silicon shortage has really helped adoption of wide bandgap semiconductors. I think the answer is maybe yes for the cheaper gallium nitride, probably not for silicon carbide. Silicon carbide is arguably in bigger shortage than silicon.
And also I would not get excited about the new shiny things like diamond and gallium oxide. In this field it takes a lot of time to bring a new thing online. Customers tend to have very stringent reliability requirements. And bringing production to the quality and profitability levels of silicon carbide and gallium nitride will require a lot of investment over many years.