Today we are pleased to launch a new series on the Foundry Files featuring commentary from David Lammers, a veteran reporter who has worked with the Associated Press, EE Times, Semiconductor International, and most currently, as a freelance journalist for various industry publications.
I can’t think of a more interesting topic to begin this blog series with than GLOBALFOUNDRIES’ plans for automotive ICs. Tomorrow’s cars are pulling in the need for three technologies: much faster processors based on 22nm fully depleted SOI; MRAM embedded memory; and 5G wireless communications.
Any one of these three changes–FD-SOI, MRAM, and 5G–should be enough to get the blood moving faster, but to bring them together is as big a story as the low-power application processors that emanated from the smart phone revolution starting 20 years ago.
And there is some urgency here, because ultra-fast image processing is essential to the adoption of advanced driver assistance systems, or ADAS. After 2020, autos will have as many as five cameras per vehicle, and the car’s image processors must be fast enough to react instantaneously to anything in the path of the car.
Let’s take them one at a time, beginning with the arguments for why GF is committed to FD-SOI at the 22nm node for automotive-use MCUs.
FD-SOI excels in two areas: since junction leakage is suppressed by the buried oxide layer, power consumption is constrained, making it easier to meet the temperature requirements of automotive MCUs. And secondly, FD-SOI brings benefits to the radio frequency (RF) circuits in terms of linearity and insertion loss.
Jeff Darrow, automotive marketing director at GF, points out that automotive MCUs must be able to operate reliably at 125-150 degrees Centigrade ambient, with junction temperatures that range even higher. For automotive MCUs made in a 55nm bulk silicon technology, leakage already accounts for 30 percent of total power consumption.
“For bulk CMOS, leakage increases exponentially with temperature. We have to live with 30 percent leakage at 55nm, but that trend was unsustainable. We see 22nm FD-SOI providing both the low power of FDSOI with the digital shrink provided by 22nm technology,” Darrow said.
And yes, the much-improved leakage of high-k dielectrics also will be required for any automotive technology solutions at 28nm or 22nm. GF believes its use of a gate-first high-k manufacturing flow brings advantages for automotive ICs compared with the replacement gate, or gate last, approach of other foundries.
“When our competitors try to integrate an embedded Flash memory with a gate-last high-k, our analysis is that the production implementation is extraordinarily difficult. The yields would be horrendous; by our estimate, less than 50 percent,” Darrow said.
GF announced its planar 22nm FD-SOI technology in July 2015, calling it 22FDX®, and Darrow emp hasizes that “22FDX is a core part of our automotive strategy.”
With the infotainment systems inside the cabin as a separate category, the vast majority of the automotive products made by suppliers such as Bosch, Continental, Delphi, and Denso are for power train, body, and safety systems.
“What we are doing is critical for the industry, and our customers are absolutely relying on us,” Darrow said. Because of GF’s experience in making SOI-based processors for AMD and others, it has a head start in terms of SOI manufacturing know-how. Having a major fab in earthquake-resistant Dresden, Germany is a big plus as well, especially for the German carmakers, he added.
Emerging memories are also essential for future automotive processors. Today, a typical automotive MCU will have 2 MB of embedded flash, and high-end solutions can have as much as 10 MB on-board. The memory works best when it is embedded on the processor die, partly to provide the instantaneous response times, and partly to shield against RF and other radiated emissions.
Embedded flash will continue to be widely used, even as emerging memory technologies are increasingly used by SOC designers. While flash’s reliability is well-proven, it is costly to produce, requiring about a dozen additional mask layers. At GF, e-flash is being extended to the 28nm node, but beyond that the foundry is committed to magnetic resistive random access memory (MRAM) for embedded processors made for a variety of applications, including automotive.
Dave Eggleston, vice president of embedded memory at GF, notes that the semiconductor industry has “a lot of history in e-flash; it retains data well in very harsh environments. But one of our key takeaway messages is that we believe e-flash scaling is going to stop below 28. It will continue through 28 nanometers but below 28 we need a new solution, and we believe the industry is coalescing around MRAM.”
Starting with IoT solutions, and extending to storage and compute, MRAM already is being embraced by key automotive suppliers which value its power efficiency and cost advantages. And while e-flash typically requires higher voltages to write information, MRAM does not; it can run directly off of logic power.
GF has a long-term relationship with MRAM technology supplier Everspin Technologies (Chandler, Ariz.), and the partners have converged on a perpendicular spin-torque version of MRAM that has much better power consumption and write speeds than earlier MRAM bit cells.
“MRAM is a big transition. But for us it is not a question mark. We have placed our bet. We know what that next embedded memory technology is, and we are educating our customers on how that technology improves their systems,” Eggleston said.
The cost effectiveness of MRAM comes because it can be built within the back-end-of-the-line (BEOL) interconnect layers of the chip. While new deposition and etch techniques are being perfected to deal with the complex material stack of the magnetic tunnel junction, Eggleston said MRAM can be added with just three additional mask layers.
It is only in recent years that the association between cars and RF–from Bluetooth in the cabin to automotive radar to help drivers safely change lanes–has become prevalent.
Peter Rabbeni, senior director of RF business development at GF, said the 5G cellular standard was designed with automotive applications in mind, particularly the need to “see” what is around the car with latencies in the range of a millisecond.
“To make autonomous vehicles a reality requires some pretty sophisticated communications systems,” Rabbeni said, shortly after returning from the Mobile World Congress 2016 held in late February in Barcelona, Spain. The 5G standard, which was a center of discussion in Barcelona, is expected to deliver “much higher bandwidth, much shorter latencies, and support for multiple, simultaneous users,” he added.
For a crash avoidance system to make the right decisions, very high data rates and much wider bandwidths are essential. In the not-too-distant future, vehicles will be “transferring a lot of data and acting on that data very quickly, which depends on very low latencies,” he said.
Range sensing and object detection capabilities on all sides of a car are at the heart of driver assistance systems. The ADAS systems will require what Rabbeni calls “an expansion beyond 6 GHz, into millimeter wave radar, something the military has been using for many years.”
Faster data rates depend on more radios, and more digital signal processing, which drives the need for linewidth scaling. Rabbeni argues that keeping within the power envelope of automotive MCUs with RF on-board “is where things like FD-SOI have an advantage. We can leverage back-gate biasing technology to optimize the power versus performance of the device.”
FD-SOI Body-Biasing Enables Power/Performance Trade-Off and Tuning of RF/Analog Parameters
For ADAS to work, Rabbeni said “we need more complex radios to drive higher performance. We are working very hard to develop a new generation of offerings, with higher linearity, lower insertion loss, and better harmonics, which all contribute to a figure of merit for a given radio.”
When GF acquired IBM’s microelectronics operation (which essentially created the RF SOI and SiGe markets) it gained expertise and manufacturing capacity for the RF SOI-based switch and antenna tuning segments. It also gained a silicon-germanium technology widely used in Wi-Fi power amplifiers, microwave wireless backhaul and automotive radar front end solutions.
Due to the growth in wireless, the demand for GF RF technologies continues to grow and the company continues to invest in additional capacity in order to satisfy the growing demand for its technologies. While the RF SOI technologies will be built out of Burlington, VT and Singapore, the 22 nm FD-SOI products will be built in Dresden.
“We are actively working on advanced node RF SOI for next generation systems including 45nm and 22nm. The 22nm FD-SOI platform was architected with RF in mind from the start and products with embedded RF have already been taped out; test structures have been modeled and measured to further enhance the process development kits (PDKs) so customers can design in it reliably” Rabbeni said. “We have models of focused RF blocks, switches, and PLLs, to prove out how the technology can be used. We are very excited about this technology and continue to move forward.”