Enabling Automotive Design

Not all IP is created equal, and not all of it will work in automotive designs.

popularity

Falling automotive electronics prices, propelled by advances in chip manufacturing and innovations on the design side, are driving a whole new level of demand across the automotive industry.

Innovations that were introduced at the luxury end of the car market over the past couple years already are being implemented in more standard vehicles. The single biggest driver of change in the automotive sectors is electronics, and that trend is expected to continue for the foreseeable future as carmakers push into ever-greater levels of assisted and autonomous driving.

“As more and more affordable electronic equipment trickles down into entry-level vehicles, it is fueling an increase in the demand for in-car electronic devices for numerous applications such as entertainment, navigation, the monitoring of a number of conditions, and for accident prevention,” said Ranjit Adhikary, vice president of marketing at ClioSoft.

The fastest way to build chips for this market is to leverage what already has been created for other chips. But IP used in automotive applications is totally different from consumer applications. It has to function under harsh operating conditions with an expected lifespan of 10 to 15 years. Moreover, IP in vehicles has to adhere to a set of stringent but still evolving regulations, such as the Functional Safety for Automotive standards as defined by ISO 26262.

“Your 55-inch flat screen LED TV in your living room is unlikely to be subjected to a room temperature range outside of 15° to 25° C,” said Tom Wong, director of business development, IP Group at Cadence. “But an infotainment system that resides inside the cabin of an automobile, under normal operating conditions, may see temperatures in the range of 0° C to perhaps 35° C—unless you happen to take your car to Alaska in the winter or to the Mojave desert in the summer. Other automotive applications, such as the ECU that resides in the engine compartment, will see significantly higher temperatures. There is a range of different temperature classifications defined by AEC-Q100 to provide guidance to chip makers for reliability testing and qualification.”

Further, with the emergence of ADAS and autonomous driving, functional safety compliance will become a requirement for automotive SoCs and IP, he said, which is where the ISO 26262 functional standard and its related ASIL levels are relevant. “IP vendors increasingly are being asked to deliver IP that meets a certain level of ASIL readiness. Most of the Tier 1 IP vendors and foundries are responding to this challenge and requirement, and they are delivering SoCs and IP that are ASIL-B compliant. More stringent systems may have to support ASIL-C/D readiness, as well.”

Different IP emerges
Until recently, third-party IP was chosen purely on the basis of function and power/performance characteristics, regardless of what market segment it was used in. Times have changed. The automotive industry now has multiple standards for automotive IP, and the standards are being updated regularly.

“Today, one key area where automotive IP and SoC requirements diverge significantly from commercial and consumer products is safety, and this is especially important for emerging autonomous vehicle applications,” said Neil Stroud, director of technology strategy for Arm’s Embedded and Automotive business unit. “These types of products must be developed following stringent systematic process flows, which are often defined by industry standards, including ISO 26262.”

Within the car, there are multiple applications and domains, and each is addressed by different chips, as well as the IP that goes in those chips. These typically fall into three categories of chips.

“First, is the category of more traditional MCUs, which are used widely throughout the car,” said Ron DiGiuseppe, senior strategic marketing manager for Synopsys’ Solutions Group. “They are used for applications such as powertrain management, body and chassis control, operating windows, airbag deployment, and these are manufactured on traditional, mature processes. The applications processor is not running as high performance. It may be 300 or 400 MHz, and there are a lot of existing suppliers there and not as much innovation, generally speaking.”

Second are the infotainment chips. “Based on the high amount of multimedia in the car, especially video channels, some customers are deploying chips that process up to 12 high-definition video channels from different applications,” DiGiuseppe said. “That class of infotainment chip is a lot higher performance, manufactured mostly at 28nm, but moving to 16/14nm finFET-class devices. These chips include a high amount of multimedia data, video and processing. They must still be automotive grade, but because they are inside the vehicle cabin, they don’t have to meet as high temperature grade requirements as those ECUs in the engine doing power train control. There are different requirements, especially temperature and reliability requirements.”

Third is the new class of ADAS processors. “This is a new class of devices that allows the vehicle to progress between different levels of automation, and it’s where a lot of new development is focused,” he said. “When you talk about what automotive IP would go into an ADAS processor, first you have to think about what those processors are doing. Generally, they are processing a lot of different types of sensor data: radar, lidar, image sensors, camera data, and because of the amount of processing for those applications, the complexity of the chip, the amount of multimedia data, and therefore the type of development is being done in 16nm/14nm class devices, even moving towards 7nm class devices. Some of the leading edge ADAS processors are heading to 7nm.”

ADAS processors also require safety-critical features, and therefore the IP building blocks of ADAS chips need to be compliant with the automotive functional safety standard ISO 26262.


Fig. 1: Self-driving concept car. Source: Mercedes Benz

Further, even in infotainment applications, long-term reliability is critical. “Those chips need to operate with a 10 to 15 year lifetime, and from an IP perspective, that needs to meet the characteristic of long-term reliability especially when it comes to electromigration: you don’t want a fault to develop based on electromigration failure, which could be very typical in a commercial chip that is force fit into an automotive application. Therefore, the IP must be designed to prevent that, and handle long term reliability requirements such as electromigration,” DiGiuseppe added.

Arm’s Stroud agreed. “An important aspect of automotive IP is to support different/higher electromigration requirements. In comparison to other well-defined requirements, electromigration requirements are not yet standardized. Rather, they are typically based on (often secret) mission profiles from the system integrators or car manufacturers, and in the absence of public standards some foundries or IP providers have defined their own mission profiles.”

The IP also has to be general enough to be re-used across a variety of applications within a car, and specific enough to do exactly what is necessary.

“In the capabilities of cores, effective IP can be crafted from general-purpose building blocks,” said Alok Sanghavi, senior product marketing manager at Achronix Semiconductor. “ADAS subsystems can be built from multiple instantiations of general-purpose RISC integer cores such as Arm Cortex, combined with DSP cores for image processing. While plenty of EDA tools exist for multi-core implementation, they are often best instantiated in an architecture that emphasizes heterogeneous computing, such as embedded FPGAs. IP for automotive always must be designed with low power in mind, and many applications for drive train and harness also must be designed for high voltage, and must call upon mixed-signal I/O with high-speed analog and significant data-conversion bit resolution.”

Buyer beware
As can be the case with hot and emerging market opportunities, not everything is up to spec, and companies that are new to the industry are struggling to develop chips that meet specs. Not all of them are succeeding, according to industry sources. But commercially available IP may provide some help.

“IP core suppliers such as Arm, Cadence and Synopsys are enabling these companies to drastically cut the development time that it takes to design, certify, and launch safety-critical SoCs by rolling out ASIL C/D (Automotive Safety Integrity Level) ready-certified IPs for licensing, where the IP’s single points of failure in the entire system is less than 3% or 1%,” said ClioSoft’s Adhikary. “Getting an ASIL C/D certification is not trivial. It can take more than a year, working with independent certification bodies. Moreover, IPs created for the automotive industry go through a more rigorous verification process to include faults triggered by silicon or software failures, and that can take as long as six to eight months more than a typical verification process.”

This has an impact on infrastructure data and IP management as well, because IP development of for automotive industry adds new challenges. “No longer can IP management systems afford to be simple repositories with simplistic attributes and categories to aid IP search,” Adhikary said. “Instead, they have to be able to track various IPs through their development stage and accumulate the much needed knowledge-base from all sections of the development team —including software teams — to serve as the single source for all data related to the IP. This includes specifications, architecture, issues, verification results, certifications, discussions and resolutions. Engineers move on, and given the long lifespan required for automotive ICs, designers need a single system where they can find all information pertaining to an IP. For example, if an issue is found, designers need to be able to browse the knowledge-base to evaluate whether a workaround exists or find relevant information to enable a speedy resolution. It also becomes critical to track the IPs and its numerous variations across the IP subsystems and SoCs they are part of so as to inform teams if any critical issue is found. IP tracking also becomes important from a licensing perspective. As companies buy third-party IPs or forge partnerships with other companies, it is more important than ever to manage licenses effectively to mitigate liabilities and manage the security of the IP data.”

The foundry foundation
At the very foundation of SoC design today is the data and support that comes from the semiconductor manufacturers themselves. For this reason, relationships between foundries and the entire semiconductor ecosystem have never been tighter.

In order to support automotive SoCs, foundries deploy a range of different processes and geometries to support the diversity of automotive applications, Cadence’s Wong said.

Automotive ICs are manufactured in bulk CMOS, including high-K/metal gate and finFETs for more advanced nodes, as well as with fully and partially depleted SOI and bipolar-CMOS-DMOS processes. Designs in bulk CMOS can range from 1µm to 250nm/180nm, all the way down to 16/14nm. PD-SOI hovers around the 180nm to 90nm range, and FD-SOI hovers around 28nm to the recently introduced 22nm. On advanced nodes in bulk CMOS, chips are being developed at 65nm, 40nm and 28nm being, with newer applications using 16/14nm. There is even discussion about moving some of the advanced ADAS chips to 10/7nm.

The challenge is mixing long-term reliability and stringent temperature requirements with brand new leading-edge processes.

“In order to meet these goals, foundries will qualify the processes at higher temperatures, introduce more robust metal stacks and varying metal thickness to support higher current density and lower resistivity, beef up via coverage requirements, and define more stringent design rules such as metal width and spacing, as well as guidelines for via redundancy,” said Wong. “All these items are aimed at providing a very robust process for automotive applications.”

To avoid problems, Wong advises customers to question their foundry on a number of issues. (See checklist below)

Stroud noted that everyone throughout the automotive supply chain is now more engaged, and classification of automotive IP is not solely the responsibility of the foundry. “This is a shift from more vertical OEMs, which owned their own foundries, IP, SoCs and even sometimes system designs,” he said. “Now, independent foundries produce wafers for automotive designs. In the past there has been little criteria other than higher temperature support. It is only recently that some foundries have instigated wider acceptance criteria in order to have IP listed in their catalogs as ‘automotive IP,’ but these requirements vary between foundries, and at the end it is the decision of the SoC designers which IP to use.”

From the foundry perspective, there is no requirement that ISO/FMEDA and MISRA guidelines be followed, Achronix’s Sanghavi noted. But chances of a design being accepted by an automobile manufacturer are much greater if there is comprehensive documentation of IP to show that design methodologies meet such guidelines. “In most cases, it is in the foundry’s best interest to document its design methodology and to follow ISO and MISRA guidelines, if only because the cores become more widely portable and comprehensible by the customer base as a result.”

Foundry design rules
Design considerations for automotive SoCs and IP imposed by the foundry can include all analog designs needing to be SPICE-ed at higher temperatures (125C Tj for AEC-Q100 Grade 2 and 150Tj for AEC-Q100 Grade 1) because automotive-grade IP operates at a higher temperature range than consumer ICs, noted Cadence’s Wong. “All standard cell libraries will require timing models that support the higher and wider temperature ranges. Timing closure will be more difficult. Reliability analysis such as EM/IR and aging analysis for long-term reliability needs to be performed. Automotive design guidelines must be followed. On the functional safety side, the designs must be analyzed based on guidelines defined in ISO 26262. Safety manual and FMEDA reports need to be provided.”

But design guidelines for the foundry will depend on the segment of the automotive network being addressed, Achronix’s Sanghavi said. “For ADAS, infotainment, and sensor hub applications, an aggressive feature size of a high-K dielectric CMOS process probably will be desired, particularly for complex SoC designs. There will be times in drivetrain applications, however, when a high-voltage biCMOS process with more mature feature sizes will be preferred. The multiplicity of applications within a modern user-driven or autonomous vehicle means that no single process technology is ideal for everything, though a leading-edge CMOS process applicable for multicore processing often is optimal.”

Designers also need to consider when dissimilar processing elements can be combined in a single architecture. “This not only lowers overall costs of implementation, but can aid in communication between primary multicore homogenous computing blocks and the coprocessors accomplishing various imaging and networking tasks,” Sanghavi said. “Each time an ADAS complex must move off-chip for a CPU call, latency is significantly increased. That strengthens the case for using eFPGA or SoC solutions.”

Processing standards
ISO 26262 mandates what is referred to as the State of the Art (SOA) process methodology.

“There are many requirements on how to document and control all evidence, starting from the interfacing with customers and suppliers to the final release documentation,” said Joe Dailey, global functional safety manager at Mentor, a Siemens Business. “There are requirements for capturing the interfaces with distributed development; the creation and management of requirements, including the traceability to their verification; controlling the configuration of design and documentation material; and the change control for all parts of the design information. One must also define the planning, specification and reports for all verification and validation efforts of design architecture and evaluation, implementation, and tests. The process and tools used to create a product must be evaluated to a specific level of confidence. For those parts of the process and tools that do not meet the highest level of confidence, that process and tool must be further qualified to be used in an ISO 26262 development.”

As part of qualify the processes for automotive SoCs, foundries carry out a variety of activity, according to Stroud. (See chart below)

Specialty analog foundry TowerJazz serves the automotive market by adapting its foundry technologies for automotive customer needs, explained Amol Kalburge, senior director of strategic marketing for the RF business unit at TowerJazz. “For example, our silicon germanium and mixed-signal/power management technologies that are used in high volume for consumer applications such as smartphones, or tablets are also offered for automotive.”  To date, the company has shipped over 500 million ICs to automotive customers.

Kalburge pointed out that some automotive customers have more robust requirements for qualification, performance, temperature tolerance. “We provide additional characterization if and when necessary for these technologies so they can meet the customers’ requirements for automotive applications. But not all automotive customers have special requirements. So in cases where there aren’t any special automotive requirements, the customer basically takes the baseline process running for commercial-grade parts and adds screening requirements at the product level. Then they are able to repurpose the products for the automotive market.”

On the design side, partnering with EDA companies is essential, he said. “We offer additional ‘robustness’ features in our PDKs that check for design robustness to flag for certain errors that otherwise would go unnoticed or unflagged in standard flows. There is a lot of custom design and layout involved in automotive industry and these additional robustness features catch these errors early in the design cycle and eventually lead to more robust designs in volume production.”

Conclusion
Engineering teams targeting automotive applications have new challenges just because they may not have designed for this market segment previously. It is still SoC design, and the IP is not so different. And at least for now, automotive requirements are agnostic to process node. But complexity is rising even at older nodes.

“While there is so much talk about 7nm, much of the foundry business for automotive is on established nodes because they are robust, proven technologies,” said Michael Buehler-Garcia, senior director of marketing at Mentor. “The key difference is the complexity of the designs. The number of power rails, the number of power islands, the voltage to the amount of IP is mind boggling. We never thought about that when we first started using what are now established processes. That’s what’s changed. While the process is being incrementally improved, the complexity of the designs being thrown at those processes is much higher.”