How difficult is it to get zero defects in automotive chips? Automotive chip trend analysis

Some new strategies are needed on the road to zero defects.

The wave of next-generation automotive chips for assisted driving and autonomous driving systems is driving the development of new methods for critical anomaly detection.

KLA-Tencor, OpTImal+ and Mentor of the Siemens subsidiary are entering or expanding their work in the anomaly testing market or related fields. Abnormal detection technology has been used in various industries for many years, and it is one of the main technologies to achieve zero defect in chip production quality. Zero defects are crucial to the automotive field.

Typically, anomaly detection itself uses hardware and statistical filtering algorithms to locate so-called anomalies. Simply put, a chip anomaly means that the chip itself may pass various standard tests, but sometimes it will exhibit dysfunction. Such chips can affect system performance or cause system failure.

Figure 1 PAT limit and limit value graphical display

There are several reasons for the appearance of an anomalous chip or a defective chip, including the potential for reliability defects. This type of defect will not be discovered when the chip is shipped, but they will be activated in some way at the application site and may eventually be reflected in the actual operating system.

To help capture such problems in the chip, the industry often uses a variety of anomaly detection methods, such as Part Average Test (PAT). In PAT, the wafer is first electrically tested, and then the hardware method and the PAT algorithm are used in combination to detect an abnormal or faulty chip that violates a specific test specification and then throw it away.

However, the PAT method is difficult to meet the demanding requirements of the automotive industry. Michael Schuldenfrei, chief technology officer at OpTImal+, said: "The use of semiconductors for automotive and other types of mission-critical devices is growing exponentially. This trend has pushed up the need for chip quality and reliability. Using PAT or part averages The anomaly detection technique of the value test method has existed for several decades as a primary means of ensuring quality and reliability. However, in many cases, they are not very effective, or the test cost is too high in preventing missed detection."

Leak detection means that the chip passed the test and left the fab. In order to avoid this, over the years, anomaly testing experts have developed new and more advanced technologies to prevent chip misses and other problems. For example, anomaly detection is usually done during the chip packaging test phase, but in a new solution, KLA-Tencor has developed a technology for testing in a fab.

Despite this, the industry still faces a number of major challenges, including:

1. As more advanced chips are used in automobiles, new advanced anomaly detection algorithms are urgently needed;

2. Anomaly detection technology must closely follow the development trend of assisted driving and automatic driving technology;

3. NVIDIA and other IC manufacturers with no abnormality testing experience are rushing into the automotive market, which means they need to improve their learning curve.

This fast-growing automotive semiconductor market faces many other challenges. In addition to the automotive market, anomaly detection is also used in medical and other fields. According to Siemens subsidiary Mentor, in general, the commercial anomaly detection software business is between $25 million and $50 million per year. Bertrand Renaud, general manager of Mentor QuanTIx, said: "This number may represent only one-third of the actual software, because many large IDM vendors have built their own proprietary tools, and their software is not counted." Currently, this Market players include KLA-Tencor, Mentor, OpTImal+ and yieldWerx.

In 2018, the growth rate of the auto market may slow down. According to IHS Markit, the total global sales of 2018 young-grade cars is expected to reach 95.9 million units, an increase of 1.5% from 2017. According to the company's data, 2017 increased by 2.4% year-on-year in 2016.

How the growth of car sales corresponds to the growth rate of the automotive semiconductor market is not fully understood. Although the current automotive chip business accounts for only about 10% of the entire semiconductor market, this does not explain the full picture, because according to IHS Markit, the value of each car's electronics will increase from $312 to $2022 in 2013. The annual growth rate of 460 US dollars is 7.1%.

"From the hundreds of controllers and other types of electronic devices a decade ago, modern cars may contain more than 3,500 semiconductor products, and the overall cost of these semiconductor devices is continuing to rise." Rob, Senior Marketing Director, KLA-Tencor Cappel said in a blog post.

A premium car has more than 7,000 chips. Chip makers are introducing 14nm and 10nm devices into high-end models, and are also developing 7nm chips for use in automobiles.

However, in the automotive sector, two factors are unchanging – reliability and quality. For commercial chips, consumers still have some tolerance for defects. However, automotive chips do not have any tolerance for defects and failures.

This is nothing new. “Like ABS systems,” said Ben Rathsack, senior technology partner at TEL. “Because of the safety, the reliability requirements of the car are always higher.”

Therefore, automotive chip manufacturers and foundries must comply with various quality standards, such as AEC-Q100, which is primarily concerned with chip failure mechanism stress testing.

Advanced Driver Assistance Systems (ADAS) and self-driving cars have more stringent reliability requirements. ADAS involves various safety features in the car, such as automatic emergency braking, lane detection and rear object warning.

For example, NXP, the world's largest auto chip maker, recently announced a high-resolution radar chip for automotive applications. Known as the MR3003 radar transceiver, the chip is a 77GHz radar device. Based on a silicon germanium (SiGe) process, the device is suitable for front-end or corner radar applications for autopilot systems that require high resolution and long range capability.

This radar technology can track thousands of targets at the same time and can sense the surrounding environment in real time, which is necessary for automatic driving at the L4/L5 level. "These types of applications place high demands on us and the chip itself. We have carefully designed the system's internal security protocols and a series of Hooks so that sensors and cars can self-diagnose in certain situations," Patrick Morgan, vice president and general manager of the ADAS modem product line, said in a recent interview. "When we started selling these chips, we had to work hard to ensure that each chip meets the specifications. We have a zero-tolerance mentality for defects. Safety is critical and no mistakes can occur."

Kamal Khouri, vice president and general manager of NXP Technologies, NXP Semiconductors added: "Everything we do here must meet very strict automotive safety and reliability standards. To ensure that all products and solutions we recommend are safe and reliable. There is a lot of work to be done."

Safety is really critical. For example, according to data provided by Optimal+, Audi has 7,000 chips in its premium cars. Assuming that the failure rate of each chip is one in a million, then Audi will have seven faulty cars for every 1,000 cars produced. If Audi makes 4,000 cars a day, it means that it produces a faulty car every hour.

As a result, the automotive industry is working hard to achieve zero defects and other quality plans, but as systems, chips, and even software become more complex, this goal is difficult to achieve.

In its latest vehicle reliability research work, JD Power conducted a statistical survey of the number of problems encountered per 100 vehicles in the past 12 months of the 2015 model and the 2017 model. It was found that the 2017 model was reliable overall. Sexuality increased by 9%, but various electronic systems still have problems. According to the survey, the audio/communication/entertainment/navigation system is still the most troublesome product category for owners, and the number of complaints from the owners is also the highest. Among them, built-in voice recognition and Bluetooth connection are the biggest problems.

These issues may be related to the use of the latest semiconductor devices, which is why anomaly detection is critical. In anomaly detection, after the wafer fab has processed the wafers, some electrical tests are first performed and then sent to the test department for evaluation.

This method can only solve a part of potential problems. "You can't test every execution path of the device, so it's impossible to cover the entire scene. However, many different tests can now be run. Sometimes the test results are not very clear. We just know that the current method is not good enough. Said Jay Rathert, senior director of strategic cooperation at KLA-Tencor.

In addition, the test may or may not reveal terrible potential reliability defects. “Potential reliability defects are defects that are exposed when they leave the fab. They are activated to some extent by the environment, including vibration, humidity, current, electromigration or heat. Over time, they May be exposed," Rathert said.

Figure 2 Random defects

In this case, why not detect these defects before they leave the fab?

Testing at the fab

According to the University of California, Berkeley, in theory, a fab that produces 50,000 wafers per month requires the following equipment:

50 scanners/steppers and wafer tracks;

10 high current ion implanters and 8 medium current ion implanters;

40 etching machines;

30 CVD tools.

In addition, the 300mm fab is an automated factory that uses a variety of automated material handling systems and wafer transfer mechanisms to process wafers in a fab in a step-by-step manner using a variety of equipment. An advanced process wafer fabrication process can have as many as 600-1000 steps, or even more, compared to fewer mature processes.

In advanced process nodes, semiconductor devices have to deal with smaller and more precise features, and as process sizes shrink, defects become more difficult to find.

Each application has its own different defect requirements. In general, consumer-oriented OEMs are less stringent in controlling defects, but in the automotive sector, chipmakers must implement more stringent control measures in their device manufacturing processes and deploy ongoing defects. improvement plan.

“There are some prerequisites (in the automotive sector),” said Wen Wenting, vice president of UMC. “You have to have a well-managed factory and well-maintained tools. Most importantly, you need a strong quality system and a high-quality philosophy that will give you the certification you need to make your car. Very complicated. In the automotive industry, quality control begins with process design and plant planning, and extends to the actual production of chips."

In fabs, inspection systems are used to locate wafer defects. In general, chip manufacturers do not inspect every wafer because it takes a long time and is costly.