Integrated circuits made with GaN HEMTs

[ZeroFunGame]

TL;DR Summary

Why do people not make integrated circuits using GaN? Since HEMTs have so much higher mobility, why have they not been leveraged for integrated circuits?

Why do people not make integrated circuits using GaN? Since HEMTs have so much higher mobility, why have they not been leveraged for integrated circuits?

 

[trurle]

Summary:: Why do people not make integrated circuits using GaN? Since HEMTs have so much higher mobility, why have they not been leveraged for integrated circuits?

Why do people not make integrated circuits using GaN? Since HEMTs have so much higher mobility, why have they not been leveraged for integrated circuits?

Industry until recently (~2016) have struggled to make GaN PMOS devices. With NMOS devices only, maximal level of integration is limited.

 

[ZeroFunGame]

Industry until recently (~2016) have struggled to make GaN PMOS devices. With NMOS devices only, maximal level of integration is limited.

What is limiting making NMOS-only logic? Most of Intel's first microprocessors were only on NMOS. Could we make a 4-bit microprocessor similar to Intel 4004 using GaN n-type HEMTs?

 

[trurle]

What is limiting making NMOS-only logic? Most of Intel's first microprocessors were only on NMOS. Could we make a 4-bit microprocessor similar to Intel 4004 using GaN n-type HEMTs?

Practical limit of NMOS only is about 300 thousands transistors per chip. Small (8 or 16 bit) processors can be done with GaN NMOS tech, but the utility will be limited to few specialized DSP applications.

 

[ZeroFunGame]

I tried to search for GaN microprocessors or commercial ICs made using GaN, but was not able to find any. I've seen papers on a GaN comparator and 101-stage RO, but what's preventing the design of a flash ADC? The propagation delay from the 101-stage RO was 0.1 ns, which seems to be very fast, and if they have demonstrated a comparator using e/d-mode DCFL, then why have companies not yet made an ADC using GaN? Is it mostly cost? Or do you think there is a greater technical challenge with higher levels of integration?

 

[Joshy]

I admit I don't know the answer, but I believe it's process related. Industry has long invested and developed so much for silicon, that it's probably cheaper and easier compared to other materials.

If I am recalling correctly or at least in the right direction, then GaN wafers are a pain to make especially in high volume ie. they have to made by a more complicated process, developed on more expensive substrates like sapphire, and the lattice constant is hard to match to grown with buffer layers like AlN.

The channel lengths I hear about GaN seem to me less impressive compared to silicon transistors, but I'm not super aware of the capabilities regarding GaN and so maybe I am mistaken. I am thinking making more per wafer or the actual wafer size might be hard to achieve.

Just my guess... I don't know the answer.

 

[trurle]

I tried to search for GaN microprocessors or commercial ICs made using GaN, but was not able to find any. I've seen papers on a GaN comparator and 101-stage RO, but what's preventing the design of a flash ADC? The propagation delay from the 101-stage RO was 0.1 ns, which seems to be very fast, and if they have demonstrated a comparator using e/d-mode DCFL, then why have companies not yet made an ADC using GaN? Is it mostly cost? Or do you think there is a greater technical challenge with higher levels of integration?

0.1ns/stage is actually not that good; modern Si CMOS routinely have 0.02ns/stage at 65nm tech scale.
Newer paper report 0.015 ns/stage with 40nm GaN tech, with is pretty comparable with the Si CMOS.
https://www.researchgate.net/publication/260589603_High-speed_501-stage_DCFL_GaN_ring_oscillator_circuits

GaN microwave devices with modern tech allows approximately doubling of speed compared to silicon, and only for specific conditions. The cost, on the other hand, is roughly 100 times higher for GaN. Therefore, i suspect it is mostly cost.
Main advantage of GaN is not baseline speed, but capability to retain good performance at elevated temperatures (up to 275C). Therefore, it is extensively used already where high power density is needed - for example, in X to Ka bands in spacecraft transmitters.