Prof. Gabriel Rebeiz: “The main advantage of what we are showing in 22FDX is that this technology is working at very low voltages, voltages that are unheard of and unseen in the RF domain.”
When silicon-on-insulator technology went into mainstream use two decades ago for microprocessors –first at IBM and then at Motorola — technologists talked about SOI’s ability to operate at lower voltages than what was possible with bulk silicon.
Now, as 5G-capable phones are in development, that ability to operate at low voltages is in the spotlight again. A research project led by Gabriel M. Rebeiz, Professor, Electrical and Computer Engineering at the University of California, San Diego has developed key 5G RF circuits – low-noise amplifiers (LNAs) and voltage-controlled oscillators (VCOs) – which operate well at voltages down to 0.3 and 0.2V, using the 22FDX® technology from GLOBALFOUNDRIES.
“The main advantage of what we are showing in 22FDX is that this technology is working at very low voltages, voltages that are unheard of and unseen in the RF domain,” Rebeiz said in an interview. “We are not talking digital, we are talking amplifiers operating at 25 GHz with 15 dB of gain at 0.3 V and 2.5 milliwatts of power consumption. We don’t know of any technology that even approaches these values of performance in terms of power consumption and low voltage operation.”
The 22FDX process brings two advantages to 5G smartphones: the low-power-consumption RF capabilities that Rebeiz and others are working on, and digital densities (>5M gates/mm2) which enable highly integrated front-end module (FEM) designs.
The 5G standard employs the 6 GHz band and two millimeter-wave bands, with 28 and 39 GHz as the two bands most designs are focused on, said Shankaran Janardhanan, director of RF offerings at GF. More than 50 wireless network operators in 66 countries are expected to roll out 5G capabilities, with the first-generation millimeter wave 5G handsets expected to be in operation in the 2020-2021 timeframe.
Janardhanan said clients focused on 5G ICs are turning to 22FDX for highly integrated designs. “The reason clients are working with us on 22FDX is the integration play. We have shown that we can do the analog-digital converter (ADC) circuits in 22FDX, but one of the most important specs for designers to get right is the LNA performance. With professor Rebeiz’s work, we now know we can integrate a very good LNA as part of the front end, in one design. That is the attraction, and the fact that it is cost effective,” Janardhanan said.
Rebeiz said the 22FDX-based LNAs are three to four times better in terms of power consumption compared to the same circuits implemented in the 45nm RF SOI process, which has much lower digital density. Also, the LNA circuits implemented in a 16nm FinFET technology, published last year at the International Solid State Circuits Conference, can operate at 0.4V, but the 22FDX circuits created at UCSD have less phase noise and the power consumption is much lower. “The power consumption of our LNAs and VCOs are better than the best FinFET-based design by a factor of two,” he said, adding that “we don’t know of any technology that even approaches these values of performance in terms of power consumption and low-voltage operation.”
Prof. Rebeiz holds the Wireless Communications Industry Endowed Chair at UCSD, and his former students are in leadership positions at several large IC vendors. When his group began its work with 22FDX, they first targeted the LNA design to 0.7 or 0.8V, but then dropped to a half of a volt and then to 0.4 V, and found the LNA was working well. “When we dropped down to 0.3 and then 0.2 V we found the design was still working well. By working well, I do not mean they were just barely working. No, they were working well, with very good gain and a very low noise figure. In fact, these are record numbers in ANY technology,” he said.
The circuits were common-source amplifiers, in a 2-stage common-source design. Another amplifier, a 2-stage cascode amplifier, operates at 0.5 V and results in 20 dB gain at 25 Ghz with 3.6 mW of power consumption. The UCSD group is also working on 94 GHz and 140 GHz amplifiers in the 22FDX technology and will be testing them in the next few months, he said.
Rebeiz is considered a pioneer in millimeter-wave design, but he was somewhat humble about the LNA design, giving most of the credit to the characteristics of the 22FDX technology. “We are good designers, and I don’t want to rain on my own parade. But we are really not doing anything special here. Honestly, I think for 28 GHz LNAs these are standard designs. We designed them to work at standard voltages, but the fact that they work down to 0.2 V is what is just amazing. These results are incredibly encouraging: we are talking about 0.16 milliwatts (of power consumption) per dB of gain for the common-source design and 0.18 milliwatts per dB of gain for the cascode design. There is no technology that we know today that can give so much gain at millimeter wave frequencies at so little power.”
My understanding is that analog circuits normally don’t scale as well as digital to low voltages, so I asked both Janardhanan and Rebeiz why the analog LNAs worked at such low Vdd’s. Janardhanan cited a much better sub-threshold slope than bulk, and very low drain-induced barrier lowering (DIBL) in such short channel devices. The devices have very low junction capacitance, he said, which leads to high linearity and low power.
Rebeiz had his own thoughts on the question. “There are three reasons why we can go to such voltages, talking from an RF perspective. The first is the very low capacitance to the substrate. That is a big deal for millimeter-wave amplifiers. Everything is fully depleted. And remember, this is achieved in a low-resistivity substrate, while 45 RF SOI is achieved in a high-resistivity substrate. This is nearly like a bulk substrate.”
Second, the 22FDX technology has a very high R (resistance) output, therefore the gmRo (instrinsic) gain of the transistor is very high. Third, “the intrinsic gm is very high, higher than bulk. Like a tsunami, it is a perfect storm.”
I started writing this blog on the day that GF’s CTO Gary Patton talked about realigning the foundry’s resources, “parking” the 7nm development effort and freeing up R&D dollars to further develop a “differentiated technology portfolio.”
“How do we differentiate and provide value to our clients, and not just provide something everybody is doing?” Patton asked.
Judging by what Prof. Rebeiz has discovered, it appears that 22FDX is a big part of the answer.