ARM: Bulk ports directly to FD-SOI
In a recent ASN posting, ARM Fellow Jean-Luc Pelloie said that bulk logic designs can be ported directly to fully-depleted (FD)-SOI for high-performing, low-power mobile apps.
ARM sees fully-depleted FD-SOI is a potential alternative to BULK 20nm. Jean-Luc addressed the question of what sort of impact a port from bulk FD-SOI would have on the design flow. His answer is: very little.
Jean-Luc’s article recaptures a presentation he gave at the recent IEEE SOI Conference. In case you missed it in ASN, here it is again:
By Jean-Luc PELLOIE, Director of SOI Technology, Fellow, ARM
Designs for low-power mobile applications in 28nm bulk benefit significantly in terms of increased performance and decreased power when ported to 20/22nm fully-depleted (FD)-SOI. From the designer’s perspective, the port is essentially direct – really no different from any standard port to a smaller geometry.
Interconnects, routing and RC parasitics are identical. Logic, memories, low- and high-voltage I/O and analog parts are handle in the same way as on bulk. Where there are differences, they are more at the device/process level, and do not pose any particular challenges to the designer at this point. This includes SPICE models, antenna effect, ESD protection, I/O and analog, and back-gate bias.
20nm FD-SOI v. 28nm Bulk
To get some clear figures on power and performance, ARM recently ported a Cortex-M0 from 28nm bulk to 20nm FD-SOI. We used the Cortex M0 implementation flow that was proven in 22nm SOI. This included:
- synthesis, place and route, and the same reduced set of standard cells for 20nm FD-SOI and 28nm bulk
- parasitics extraction for interconnects from the routed 22nm SOI M0 core (22nm SOI Back-End Of Line (BEOL) is considered to be the most representative of current bulk/FD-SOI 20nm BEOL)
- characterization of 20nm FD-SOI and 28nm bulk standard cells (typical process corner and room temperature)
- different voltages to create the corresponding .lib files that would be used for timing and power analysis of the M0 core: 0.7, 0.8, 0.9 and 1V
- timings and power were compared for the routed M0 core based on 20nm FD-SOI and 28nm bulk characterizations (.lib).
Source: ARM, 2011 IEEE SOI Conference
In any next-node port, you typically expect to get a 25% improvement in performance, but in porting from 28nm bulk to 20nm FD-SOI, FD-SOI boosted the improvement far beyond the expected 25%. At a Vdd (supply voltage) of 1.0V, we saw a 40% improvement in performance. At 0.9V, we saw 66%. For Vdd of 0.8, we saw an 80% improvement. And for Vdd of 0.7, we saw an improvement of 125%.
Power is consistently reduced by 30%, and leakage holds steady.
Remember, this is a straight port, which gives us a baseline figure. There are several powerful process and design optimization techniques that can boost those numbers even higher without significantly increasing the complication factor.
Existing design, tremendous results
The conclusions we have drawn are that:
- a standard bulk ASIC design flow can be used for FD-SOI – don’t expect any change
- an existing bulk logic design can be directly ported to FD-SOI
- you just need to check the timing closure – there is no timing variability (this is not PD-SOI)
FD-SOI should give tremendous advantages in terms of both power and performance. These low-voltage, high-performance chips are perfect for low-power applications, with the undoped channel in the low voltage SRAM resulting in higher margins. For some applications, RF features will also be improved if designers choose high-resistivity substrates.
FD-SOI is all a designer needs for high-performing, low-power mobile applications. And happily from the designer’s point of view, FD-SOI is as simple as designing in bulk.
This article was adapted from “FDSOI Design Portability from BULK at 20nm Node”, which was presented at the 2011 IEEE SOI Conference.