Chipmakers Test Ferroelectrics as a Route to Ultralow-Power Chips | Hacker News Reader

Chipmakers Test Ferroelectrics as a Route to Ultralow-Power Chips

spectrum.ieee.org

62 pointsrbanffy25 comments

Supriyo Datta is a pretty top level semiconductor theory guy, so its a good chance theoretically this could work, but the place where tons of issues have come up over the years in using new materials is that they often lead to quality or defect issues in the manufacturing process that takes years to figure out.

Fwiw, Chenming Hu is a proponent of NC-FETs as well.

Also, GloFo has already fabricated FinFET NC-FETs.

oh interesting, was not aware of the term NC-FET -- so basically the ferroelectric material gives it the negative capacitance behavior... interesting.

Yup, the NC-FET essentially has the NC and positive capacitor in series to get a voltage amplifier, a clever way of subverting the Boltzmann limit :)

Also funny, I spent months trying to buy his book for one of my semiconductor device physics courses that was out of print, and finally bought it for like $180 on ebay.. it actually was worth it.. really well written book..

Could you please name the book for those of us that may be interested?

I'm not xt00, but based on age/out-of-printness I'm guessing it may be Datta's 1989 book "Quantum Phenomena (Modular Series on Solid State Devices, Vol 8)". His 1986 book "Surface Acoustic Wave Devices" is a distant second-best possibility, depending on what was in that one semiconductor device physics course.

Everything else seems to be in print still, though some are expensive (as typical of specialized books).

Was definitely part of the modular series, a small dark blue book.

It’s this one as the previous poster said. "Quantum Phenomena (Modular Series on Solid State Devices, Vol 8)".

It’s referenced to by many other books from what I recall. So I decided to go hunting for it — and found that the explanations in Datta’s book were better than many of the books that came after it. I’ll see if I still have the book someplace.

There's also a decent scan of the book available via b-ok.org.

>Some researchers have predicted that transistors built with ferroelectrics will never exceed 100 megahertz. And some think that building these devices will require very thick layers of ferroelectrics—too thick to be practical.

Wast amounts of low power MCUs are still built to function in single MHzs. A jump from milliamps to microamps is not a small deal.

Less than 1MHz is also very common - some MCUs have the option of running from the 32kHz RTC oscillator in low-power modes.

Momentum is definitely building on this, although still very much in the R&D phase https://semiengineering.com/a-new-memory-contender/

I wonder how this is related to Ferroelectric RAM, which offers nice properties along all axes except density?

So in NCFETs there is something called capacitance matching which is basically the value of the negative capacitor compared to the value of the positive capacitor.

When you have different values, then you can get NCFETs, and in different values, you get FRAM.

In FRAM you get states stored in the polarization of the atoms in the ferroelectric layer.

Is anyone putting R&D into this besides Global Foundries? I've seen their name pop up in every spintronics announcement lately.

Probably not since there aren't enough competitors left to actually invest in chips. If you don't hear Intel, TSMC or Global Foundaries putting R&D into something then you have the way out there research of academia.

Maybe a random startup or two are investigating this but those are academics that figured something out from a grant and are attempting to make a company out of it. The road to beating silicon is a long one.

TI and Cypress through Ramtron has done quite a bit of work incorporating ferroelectric materials into CMOS processes and such for FeRAMs mostly. Obviously RAMs are not transistors but the work is certainly relevant. Like the article says, the large feature size for ferroelectrics has mostly been the limitation for ferroelectric applications.

GloFo is a major proponent of MRAM not sure how invested they are in spintronics in the general.

Sure, you can reduce the voltage required, but what about the current required? Ferric materials tend to be somewhat resistive compared to most other metals.

The ferroelectric materials here are used in the insulating stack, so actually higher DC resistance is a good thing.

The ferroelectrics here are used to create a negative capacitance, which, when in series with a positive capacitor, results in a voltage amplifier. This subverts the 60 mV/ decade Boltzmann limit.

DDR4 on chip frequencies are on the order of 200mhz internally. They get the high data rates by being massively parallel. So never discount the low clock frequencies.

It's a moot point anyways since hafnia based ferroelectrics do not suffer from this speed issue. Also this is the transistor switching speed we're talking about, entirely different than DDR clock speeds.

Transistor speed is actually the problem on DRAM memories. The process for making great capacitors is terrible for making fast transistors.

But DRAM processes use RCAT transistors.

Reply on news.ycombinator.com