Competitors in the Internet of Things space are set to describe their chip designs at the upcoming Mobile World Congress (MWC 2018) in Barcelona, Spain, including several startups using GLOBALFOUNDRIES’ 22FDX® process.
Anup Savla, chief technology officer at Nanotel Technology, said his young company is designing several chips for use in the narrow band (NB) IoT space. Savla, who spent 11 years designing wireless ICs at Qualcomm after a three-year stint at Intel, said Nanotel chose to use the 22FDX process to reduce power consumption for its mixed-signal NB-IoT modem.
“We have a digital engine, a processor, designed around IoT applications, where the emphasis is on low power and low leakage. With 22FDX there are knobs that are available to turn down the power and leakage. The opportunities to do that are unparalleled, and you just don’t get that kind of opportunity from bulk CMOS,” he said.
The Nanotel transceiver design started before the first 22FDX design kits were officially released, using a 0.5 PDK, but with libraries targeted to the 0.4 V operating voltage. “From the beginning we were definitely targeting the .4 V library. The reason is that at the .8 V level you are not differentiating on power enough, relative to a bulk CMOS process. It is at .4 Volts where you are really getting power consumption levels that are significantly lower, at similar costs to a bulk CMOS process,” he said.
Asked about designing to an FD-SOI process, Savla said “we just had typical early adopter-type issues. Part of it was the additional modeling and testing requirements to use the back gate, but that will remain true regardless of the maturity of the process. It is not necessarily a negative. If you really want to exploit what this process can do, then you want to vary the back-gate voltage, but that comes with the added modeling.”
Savla said the Nanotel chipset design is divided equally between digital and analog. “Within the same design we can use switches with backgate control to cut off the leakage current when the device is basically sleeping and not in use, to an extent that is not easily possible in bulk CMOS. On the other hand we can use the back-gate in active mode devices to make active operation possible with very, very low supply.”
Nanotel’s primary focus is not to sell ICs – it is a solution-focused company, designing devices and data packages for its customers, allowing them to use long range, low cost connectivity to the cellular network without having to rely on WiFi. Having its own chipset gives Nanotel a means to reduce costs and customize unique features for its customers, he said.
Dual-Mode Connectivity Solutions
The leading low-power wide-area (LPWA) connectivity standards — LTE-M, which is gaining traction in the U.S. market, and Narrowband IoT (NB-IoT), which being adopted in Europe and Asia – are expected to boost IoT deployments to nearly half a billion by 2021, according to ABI Research.
GF and VeriSilicon are developing a suite of IP to enable customers to create single-chip LPWA solutions that support either LTE-M or NB-IoT, a dual-mode solution. The IP enables a complete cellular modem module on a single chip, including integrated baseband, power management, RF radio and front-end components.
VeriSilicon provides Silicon Platform as a Service (SiPaaS) intellectual property which allows its customers to focus on differentiating features. VeriSilicon CEO Wayne Dai said the Chinese government has targeted NB-IoT for nationwide deployment over the coming year. GF’s new 300 mm fab for FDX in Chengdu, and IP platforms such as the single-chip solution for integrated NB-IoT and LTE-M, “will have a significant impact on China’s IoT and AIoT (AI of Things) industries.”
One Picoamp Per Micron
Anubhav Gupta, director of strategic marketing and business development for IoT, AI & Machine Learning at GF, said some customers are taking older multi-chip designs and creating single-chip solutions in 22FDX. “Due to the efficient SOI FET stacking for high power PA and high switch linearity, there are area, power and cost advantages in moving to a single die in 22FDX. We see low short channel effect, higher transconductance gain, significantly better mismatch and lower noise than equivalent designs in 28nm bulk.”
On the digital side, Gupta said the body-biasing capability allows customer to operate as low as 0.4V with standby leakage currents of less than one picoamp per micron. Also, GF now offers embedded MRAM with very fast wakeup, a similar read speed to flash, but a 1000x better write speed. When eMRAM is used in combination with on-chip SRAM, customers can avoid off-chip flash completely, Gupta said.
A Roadmap Required
Dan Hutcheson, CEO of Santa Clara, Calif.-based market research firm VLSI Research Inc., conducted a survey in 2016 of designers, asking for their opinions on fully depleted silicon-on-insulator (FD-SOI) compared with bulk CMOS. “The problem, people said back then, was that they didn’t know if there was a roadmap. Since then, GF has developed the 12FDX™ platform to follow on from the 22FDX, so that issue should be off the table.”
Hutcheson said he believes there are designs underway by companies that are holding their cards close to the vest. “Since 2016, there is a lot more IP available, and GF has addressed the issue of the roadmap with 12FDX, so that 22 is not just a one-stop thing.”
A spokesperson from STMicroelectronics, said “since 22FDX integrates the second-generation active devices (transistors), it was natural for ST to select GF’s 22FDX technology as our next-node technology, after the 28nm FD-SOI technology we are already using.”
The spokesperson said ST takes a positive view of the “development of the 22FDX technology node in Dresden, which is now qualified for volume production and ready for primetime, so it is immediately usable by ST to develop products.” The wafer capacity and experience of the manufacturing team in Dresden “give us confidence in GF’s capability to qualify and produce in volume.”
‘Performance-Optimized’ Vision Processor
Jens Benndorf, chief operating officer of Dream Chip Technologies GmbH (Hannover, Germany), said his team used 0.8V libraries for its “performance-optimized” automotive vision processor. Dream Chip was the lead company in an EU-supported design project that included ARM’s A53 Quad and Cortex®-R5 lock step for functional safety, Cadence’s quad Vision P6 , FlexNOC from Arteris IP, LPDDR4 controller from INVECAS, and other IP partners. The resulting multi-core vision processor, based on the 22FDX process, was first unveiled a year ago at the 2017 MWC. Since then, the design is providing European auto makers and Tier 1 automotive component suppliers with a platform from which they can create custom derivatives.
“The automotive industry realized that their assisted driving solutions, besides Radar and Lidar, required more camera information, integrating information from multiple cameras. The resulting Multi-Processor Chip Design used forward biasing to boost performance, and not any back biasing,” Benndorf said. The result was a computer vision processor solution, measuring 64 sq. mm, with an estimated 1 billion transistors, and drawing 4 Watts, which he said is “a very aggressive power consumption number given how much vision processing is on the chip.”
Riot Micro Bets on Cellular Links
While Nanotel’s design is equally divided between digital and analog, another startup using the 22FDX process has an all-digital NB-IoT modem design. Peter Wong, CEO of Vancouver-based Riot Micro, said his company’s approach, which does not use a digital signal processing (DSP) approach, allows IoT customers to turn off large portions of the chip to save power. That is especially welcome for battery-powered IoT edge devices that might need to operate for a decade on a battery.
Riot Micro’s first design was done with a competing foundry’s 55nm bulk CMOS, but a follow-on chip is in the 22FDX process. The Riot Micro LTE Cat-M/NB-IoT modem includes an ultra-low power processor to run the protocol stack. “We borrowed design methodologies from the Bluetooth world to drive down the power and cost. The PHY is designed using gates instead of a DSP with a tightly coupled and highly optimized prototol stack, this gives us very fine granular power control over the modem.“, Wong said.
“With 22FDX, the value proposition for us is potential power and area savings,” said Wong. “In addition, leveraging the growing ecosystems of IP availability in the 22FDX process helps to accelerate time to market.”
The Riot Micro design is a digital cellular modem which supports the LTE Cat-M and NB-IOT cellular standards; Wong said the Riot Micro modem will be certified with several major cellular carriers this year. A customer in the Middle East is planning to use it for an emergency-alert system.
“There are many ways to connect things to the internet: WiFi, Bluetooth, Zigbee, cellular, etc… and there are use cases that fit all of them, but for many applications, cellular has so many advantages. Cellular is inherently more secure, easy to deploy, provides mobility, and the spectrum is licensed and managed. Just turn it on and it connects. You don’t have to worry about spectrum; that is all managed by the carrier.” he said, citing asset trackers and asset management as key applications.
Integrated Power Management
Gupta said GF sees some mixed-signal IoT customers trending toward a 0.4V power supply for the digital circuits, and 0.8 to 1.8 volts for the analog portions. “The availability of LDMOS in 22FDX removes the requirements for an external PMU (power management unit) for low power IoT applications. Typically in bulk processes they don’t have high voltage LDMOS, and since a lot of the IoT applications work on lithium-ion batteries, these applications would require an external power conversion chip for battery-powered applications.”
And the 0.4 V designs have enough digital performance to support an ARM core, for example, running from 100 Mhz up to sub-500 MHz speeds, Gupta said.
Tim Dry, segment marketing director at GF, said engineers are beginning to more fully understand the analog design capabilities of the 22FDX technology by using dynamic body-biasing. “It turns out that SOI body biasing enables of lot of analog scaling that we didn’t understand until recently. For ADCs (analog to digital converters), radios, and power components, we believe we can get the die area much smaller than with existing planar and potentially FinFET.”
The 22FDX solutions for IoT systems, such as smart meters, augmented reality and virtual reality headsets, utility control, and security cameras, can reduce power consumption. “Smart speakers are another application getting a lot of attention,” Dry said.
For more information on GF’s FDX™ solutions, join us at MWC from February 27 – March 2 at the Fira Gran Via in Barcelona, Spain, to learn about how GF’s technology platforms are positioned to enable a new era of ‘connected intelligence’ with the transition to 5G, or go to globalfoundries.com.