GF Ready with a ‘Body Bias Ecosystem’
By Dave Lammers
GF has had some success with the 22FDX technology, but in the conservative semiconductor industry, getting a new technology up-and-running takes time and effort. After several years of laying the groundwork, the time is now for FDX, as 5G, automotive, and RF-enabled IoT devices all represent fast-growing markets for the fully depleted SOI technology.
Ed Kaste, vice president of Industrial and Multi-Market at GF, told a GLOBALFOUNDRIES Technology Conference (GTC 2019) breakout session that GF expects to ship about 100 million 22FDX devices this year.
One volume design win came in the power management space, where FDX’s advantages in analog, RF and low-power operation provided an edge over bulk devices. Jamie Schaeffer, a senior director of Mobility and Wireless Infrastructure Market at GF, said, “we believe that showing what FDX can do in that power management application will help us going forward to gain other high-volume design wins in the IoT space. “While the power management die are relatively small and the number of wafers required is not huge, Schaeffer said it is an example of how 22FDX – even without body biasing coming into play -- has shown that it “is simply a better transistor.” At .5V, which is where many of the design wins have been targeted, the SOI-based FDX technology consumes much less power, has far less leakage, and the RF performance has “exceeded even GF’s initial expectations,” Schaeffer said at GTC 2019, held in Santa Clara in late September.
The sense I got at the conference is that it is just now that all the pieces are coming together for the FDX technology – more pieces than can be described in this one blog, so I’d like to focus mainly on the adaptive body bias capability.
Very few of the FDX design wins have used body bias, partly because not many designers have heretofore been able to readily implement body biasing to improve performance or reduce power consumption. And in some use cases, the IP that was used to implement body biasing itself consumed a fair amount of power, negating some of the overall gains in power consumption.
Body Bias Ecosystem Ready
But all that is changing. Adaptive body biasing (ABB) is now fully supported, and overall a much deeper FDX IP library is available from several IP vendors, including Synopsys for automotive-use applications, ARM for IoT CryptoIsland™security, IP from VeriSilicon, and several others.
Schaeffer told the GTC participants “body biasing is arguably the most differentiating capability of FDX, but it has not been widely adopted because the complete body-biasing ecosystem has not been there. Over the last couple of years, we have built the first body-bias ecosystem, including the libraries and memories with process, voltage, temperature, and body bias characterization. There are body bias generators, and reference designs from Cadence and Synopsys. We are ready, from design and planning all the way through characterization and test.”
Schaeffer said the goal is to “simplify the use cases for our customers. There is no limit to what a creative designer can do with body biasing, but to some extent we overcomplicated it. To simplify the design cycle, we are enabling our customers to compensate for the process with body biasing. That alone will provide for up to a 30 percent improvement in the performance of the design. When you think about it, that is similar to a half-node or full-node improvement with just a design tweak.”
Schaeffer announced at GTC that ARM’s built-in security IP, called integrated Secure Element (iSE), is now supported on the 22FDX platform, making it easier for IoT device designers to gain security certification.
ARM’s integrated Secure Element (iSE) IP, now available for 22FDX-based SoCs, lowers the barrier for entry for silicon vendors requiring high-security standards
(Source: GTC 2019)
Key IP from Dolphin Design
Philippe Berger, the CEO of Dolphin Design (Meylan, France), said 30 of his engineers this year developed IP and test-chips that support adaptive (forward or reverse) body biasing (ABB) on the FDX platform (a forward bias lowers the threshold voltage to boost performance, while a backward bias raises the Vt to reduce leakage current).
The ABB feature allows designers to leverage forward and reverse body bias techniques to dynamically compensate for process, supply voltage, temperature (PVT) variations and aging effects. Dolphin’s IP embeds the body bias voltage regulation, PVT monitors and aging sensors, and a control loop. As the sensors monitor the PVT characteristics, “the bulk voltage can be adjusted to bring back the transistor characteristics to where you want them to be,” Berger said.
The Dolphin IP, he said, “provides a design methodology to fully leverage the benefits of corner tightening.” Dolphin signed a contract to develop the ABB intellectual property with GF in January of this year, and the IP is now available. While the customer licenses the IP from Dolphin, Berger said it is free of charge as “GF-sponsored” IP.
“Basically, almost no one is using body biasing today, because the design environment was not there,” Berger said, predicting that “going forward, body biasing solutions will be important, particularly as designers incorporate RF with the rest of the circuitry.” The BodyBiasing capabilities arrive as the 5G cellular and IoT markets blossom, Berger said, with BB used widely in the power amplifier (PA) and low-noise amplifier (LNA) functions.
Dolphin’s adaptive body bias (ABB) IP is part of GF’s body-bias ecosystem, designed to improve performance, latency, and energy efficiency.
Source: GTC 2019)
Bhaskar Kolla, vice president of customer engineering at Invecas, Inc., said Invecas has created a standard cell library with the right corner characterizations for applying adaptive body biasing. The “hooks” in the Invecas libraries work together with the ABB IP from Dolphin. “The Dolphin IP will make it easier to use dynamic (adaptive) biasing. Some customers have used static (forward) biasing, but with adaptive biasing the customers can use either forward or backward biasing. Our libraries now support that,” Kolla said.
John Pellerin, chief technologist, Platforms and vice president, Worldwide R&D at GF, said there are further ABB advantages coming in the next-generation FDX process, 12FDX. The well structures in the 22FDX process require that the body bias condition for a given circuit block to be pre-determined, plus or minus from nominal voltage, while with the 12FDX process it will be possible to switch the bias in either direction at a very fine granularity of circuitry. “I like to compare it to a street. With 22FDX it is like a city block where traffic can only go in one direction, while with 12FDX traffic within that city block can go in both directions.”
GF has reorganized, creating one division that handles the process platforms, IP, and other design-enablement elements for Automotive, Industrial, and Multi-Market (AIM) applications, the latter encompassing IoT-type markets such as low-power cellular networks used in agriculture and smart cities, medical, wearable electronics, smart homes, and several others.
Kaste told the GTC audience that despite a pullback from earlier hype about the growth of the IoT market, he believes “we hope to demonstrate that it is absolutely taking off now,” with the AIM markets of automotive, industrial, and multi-market growing to 24 billion units in 2022. Moreover, nearly all of that AIM market (98 percent) will use design rules of 12nm and larger “until 2022 and beyond.”
Every Watt Counts
To make those predictions become reality, limiting energy consumption is paramount, representing a great opportunity for the FDX technology. The billions or even trillions of connected devices must be drawing extremely low power. As Synopsys co-CEO Aart de Geus noted in his keynote address at GTC, in an era of global warming, “every Watt counts.”
The combination of 5G networks, enabling AI-enhanced IoT devices to send much larger amounts of data to the cloud, represents a multiplier effect. “IoT, multiplied by 5G, will create immense amounts of data being sent to the cloud. But that represents unbelievable amounts of energy demand, coming at a time when everyone must be concerned about climate change. The very survivability of the planet is at stake,” de Geus said.