Assembly houses are fine-tuning their methodologies and processes for automotive ICs, optimizing everything from inspection and metrology to data management in order to prevent escapes and reduce the number of costly returns.
Today, assembly defects account for between 12% and 15% of semiconductor customer returns in the automotive chip market. As component counts in vehicles climb from the hundreds to the thousands, and quality targets shift from 10 dppm to 10 dppb, assembly engineers need to find practical means of delivering zero defective parts. Doing so puts greater demands on various process steps, including, metrology, inspection and test.
While semiconductor test engineers are making great strides on isolating fab-generated defects, assembly engineers are quietly focusing attention on improving inspection and processing of equipment data to catch latent defects. This is a big deal for automotive electronics. According to a BMW presentation at the 2017 Automotive Electronics Council reliability workshop, most semiconductor devices fail within the car’s warranty period.
The carmaker noted that 22% of warranty costs are due to electronics and electrical control units. Of those failed parts, BMW said 77% of the failures are semiconductor devices, and 23% of the parts are isolated to active and passive components. Of those semiconductor failures, 48% were due to systematic fails, 24% to test coverage, 15% to random failures, and 6% were retested and did not fail the second time. The failure pareto was also broken down to 41% final test, 24% front-end processing, 22% design, and 12% assembly.
Historically, assembly factories have limited investment in process control and 100% inspection steps, partly due to lower process complexity and partly in order to maximize profitability. While final test can detect opens and shorts it is less effective at detecting marginal assembly defects that lead to reliability failures. Now, in order to meet the zero defects goal, assembly houses are ramping up 100% inspection capabilities, and investing in process control at key assembly manufacturing steps.
“In the automotive segment you cannot reach that level (10 defective ppb) without doing 100% inspection,” said Frank Chen, director of applications and product management at Bruker. “Reliability-related defects may not be caught by either electrical test or lifetime test. The big emphasis is on expense, feasibility, and scalability.”
Fig. 1: Impact of 100% inspection on assembly process. Source: Bruker
Sean Langbridge, European sales director at CyberOptics, pointed to a number of concerns for automotive customers when considering an inspection solution. Among them:
- 100% defect coverage = no escapes
- Extremely low false failure rate
- Simple actionable process improvement data
- Reliability / Uptime
- Minimum Engineer overhead = Programming & tuning time
As each new fab and assembly technology enters the automotive sector, the challenge is to understand what defects need to be detected and what equipment parameters may impact defects. Learning from the feedback of downstream steps requires data, as well as a means to connect the data. Unlike wafer fabs, this traditionally has not been an area of high investment for OSATs.
“There are two important activities in order to have assembly facilitates help in increasing overall device quality,” said David Park, vice president of marketing at PDF Solutions. “First, you have to be able to collect data. As the saying goes, if you don’t measure it, you can’t improve it. When you invariably receive an RMA, if you did not collect manufacturing data, you don’t have a way to get to root cause. And to leverage that data that you need single device traceability capability through the back end, i.e., manufacturing through the assembly and packaging process.”
Automakers are demanding 10 defective ppb limits across the whole spectrum of automotive ICs and single transistor/diode devices. Depending upon the semiconductor device type assembly processes differ. This article focuses on ICs that use wire bond and bump technologies.
Assembly processes and associated defects
Automotive semiconductor products cover a wide range of process technologies, package technologies, and device sizes. Device can be as small as a single power FET, or as large as 2 billion gates in a custom SoC used for processing sensor data from cameras and lidar. Typically, automakers use well-established semiconductor and assembly technologies, which provide the benefits of predictable quality and reliability levels. With a vehicle’s increase in computing and communication capabilities, automakers have upgraded to more recent technologies. In fact, they are now using finFETs in microcontrollers.
Fig. 2: Variety of packaging technologies used in automotive. Source: Amkor
As automakers shift to more advanced process nodes, the package technology shifts from wire bond (WB) to BGA and FCBGA. For instance, at 64nm and 40nm, CMOS WB packages predominate, but BGA is not uncommon. Defects associated with wire bond technology include opens, shorts, marginal bonds, dirt-on-lead, die cracking, die adhesion to substrate, mold cracks, foreign materials, and package delamination.
From 28nm to 7nm, FCBGA is the predominate package technology because it offers the pin density required by larger SoCs. These packages introduce new assembly defects that may pass final test yet fail in the field. New defects involve white bumps, bump cracks, and bump-to-substrate joints, all of which have a reliability sensitivity that makes them high risk for non-detection at final test and failures in the field.
With their foray into newer assembly and fab processes, automakers have well-acknowledged trepidation. First, vehicles easily have a lifetime of 10 to 15 years, as opposed to the 2 years for consumer products, and 4 to 7 years for computers. Understanding of long-term reliability issues is largely theoretical. Second, these processes have yet to experience the harsh automotive environment, which can translate into varying long-term impacts for marginal defects.
A key premise for die entering the assembly process is that they are good die. Yet assembly processes can damage the die, impacting quality and reliability.
Understanding the assembly process informs one as to where and when assembly-related defects are introduced. Conceptually, wire bond and BGA/FCBGA assembly processes need to singulate the die, attach it to a substrate, bond the die to the substrate, test the unit, and perform a final inspection.
Fig. 3: Wire bond assembly process. Source: PDF Solutions
Within each process, manufacturing defects can occur. Meeting automotive expectation of flawless products requires an investment in 100% inspection capabilities, and more advanced tests at final test. Broken contacts and shorts between contacts/wires are easy to electrically test. Marginal contacts, die cracks and delaminations (bumps/package substrate) all represent latent defects that will fail in a vehicle. Marginal contacts can pass electrical test, yet because of a poor intermetallic connection, subsequent rapid thermal cycling and vibrations will cause the connection to open at a later point in time.
PCB manufacturing and assembly shares similar concerns with marginal contacts. In automotive electronic units, High Density Interconnect (HDI) packages are becoming more popular. As with the assembly process, there are concerns with marginal contacts that escape electrical test. PCBs in assembly have similar challenges.
“Several automotive PCB experts recently told us that a significant part of the quality issues that they currently deal with are due to HDI boards malfunctioning as a result of bad inter-layer connections through the laser vias. This is occurring, despite the fact that the PCBs had successfully passed all the electrical tests prior to shipment,” said Micha Perlman, product marketing manager for via formation, Orbotech, at KLA. “HDI involves laser drilling and plating for inter-layer connections. Any residues or contamination will cause bad connections between the layers through the copper plating process. These defects cannot usually be detected after plating and therefore may pass the PCB’s final electrical test. However, due to the high stress of the component assembly process or at some point during the end-device’s operation, latent defects introduce high risk for future failure. This is why laser via inspection — which was relatively rare a decade ago — is more common today and is in high demand from car electronics manufacturers and other products that require high reliability and safety.”
The sawing of die from the wafer represents the first opportunity for assembly process to introduce defects. Cracks in the die are the leading defect mechanism.
“Hidden defects may not cause device failure at the manufactured factory, but they can cause long-term reliability issues in real world usage. Sidewall cracks and inner cracks are initiated from the sawing process,” said Woo Young Han, product marketing manager at Onto Innovation. “These cracks are not easy to detect with the traditional BF and DF illumination optics from the wafer surface. These need new imaging method to detect these types of cracks. During pick-and-place, the machine picks up each die and takes images from all four sides to see if cracks can be seen. Time and cost can be greatly reduced if these cracks can be prevented or detected prior to pick and place operations.”
Fig. 4: Sidewall and inner cracks. The primary source is die singulation, i.e. wafer sawing. Source: Onto Innovation
Fig. 5: Inner cracks are difficult to see from top-side imaging. Source: Onto Innovation
“Presence of an inner crack underneath the die surface can be confirmed with the IR image,” Han said. “However, the contrast of the inner crack is very low, since it is underneath the surface and such defects are difficult to detect. Unlike the IR image, the same inner crack has over 200 GSV contrast.”
Die attach: Wire bond and BGA
Prior to a wire-bonding step, die are glued to the package substrate using epoxy. While wire bond technology has been around for decades, there is still significant process complexity.
“There’s a lot of really detailed information going on during a wire bond,” said Dave Huntley, business development director at PDF Solutions. “For example, a package might have 5 dies, and every die has 20 to 30 wires. For every individual wire there’s data associated with each bonding step (2 bonds per wire). Each one of those bonds has a list of process variables and a number of equipment generated waveforms of voltage-currents. Also, there may be an image. In one process we have looked at, there were 59 variables. We assessed we’re collecting about 3 megabytes of raw data for a single multi-chip package.”
With this data, assembly process engineers can apply FDC and SPC analysis, which are commonly used in wafer fabs, to improve the overall process. Inspection and metrology can inform on the quality of the connections, and thus provide feedback as to which process equipment parameters matter most. For wire bonds, inspection catches defects while metrology systems assess the loop height, width, lift, wire length, and position. In assembly processes, the challenge to reach 100% inspection requires high-throughput systems that can automatically and reliable identify random and systematic defects.
Assembly facilities commonly use optical inspection systems. To perform inspect wire bond loop height requires 3D imaging from multiple camera angles.
Fig. 6: 3D automated, optical inspection of wire bonds. Source: CyberOptics
X-ray technology provides another potential approach for 100% wire bond inspection. X-ray technology can be extended to view high-density bumps in the wire bond realm. “We actually have a prototype of the detection method and are qualifying it with respect to the performance requirements,” said Bruker’s Chen. “It is using more angles of views to do this because you need that reconstruction ability to see the different wire paths.”
CyberOptics’ Langbridge noted the following list of features that are difficult to inspect:
- 3D wire bond loop profile on bonds less than 80µm diameter;
- Bottom terminated devices (BTD), which mostly require X-ray or computerized X-ray (CT), and
- Subtle solder-related defects, such as poor wetting or cold solder joints.
Poor connectivity between a die bump and substrate contact is of particular concern for assembly manufacturers, because these will pass electrical test and may fail in the field due to a metallurgical connection that is not robust.
Optical inspection can detect issues with bumps on the die or substrate prior to bonding. But it has significant limitations on detecting bonding issues, which leaves X-ray technology as a possibility.
“The most critical thing we’ve heard from automotive customers is that they want to detect these latent issues, like partial wetting,” said Chen. “This is their big fear because they can catch completely open circuits. If a contact is 50% to 90% marginal, you want to catch those cases before it ships out, and they don’t have a good method to do that right now. In addition, with inspection we can provide process feedback and early detection of excursions.”
Looking ahead to hybrid bonding
While hybrid bonding (a.k.a. thermal-compression bonding) for dense package connectivity has yet to show up in an automotive IC, the inspection industry can support it by leveraging the systems built for high-performance computing ICs. X-rays can see through the die bump substrate layer.
“There are X-ray solutions that exist, but there’s really been a gap in the sweet spot of having a technology that has enough resolution or sensitivity while being fast enough to do 100% inspection,” said Chen. “Additionally, for X-ray tools, there’s a history about being careful with the X-ray dosage because it can potentially damage a part. It’s often viewed as a destructive technique for memory devices, so there’s also teaching customers that there’s a way to do X-ray inspection safely, and we can do it at high throughput and inspect every device.”
Fig. 7: X-ray inspection attributes. Source: Bruker
The detection of anomalies can provide more than a pass/fail decision.
“When using thermal compression, each bond head has a fingerprint/signature of behavior,” Chen said. “There are some common displacement issues or pressure differential for which we have process specs. We can flag when it’s really outside of the spec, where it’s generating a lot more latent defects or true hard defects. Not only can we detect those issues, but we can also provide a grade of marginality — identifying if it’s a killer or a latent defect where it’s so marginal that you don’t want to release the device.”
Final test does an excellent job at detecting gross defects such as opens and shorts. Toni Dirscherl, product manager for power and analog solutions at Advantest Europe, pointed to the typical tests that focus on assembly related properties/defects.
- Continuity test can determine if all pins are bonded correctly and have no shorts or opens.
- Thermal resistance (Rth) measurement of power devices checks for thermal distribution inside the device, indicating proper connections.
- Stress voltage tests measure the leakage current delta before and after the stress testing to see if the device has been damaged.
And like everything in automotive, keeping costs down remains a focus. So burn-in, which is typical for complex microprocessors, is not a preferred option because it doesn’t necessarily accelerate package defects. In addition, the cost of parallel test keeps coming down.
“With low margins on the majority of automotive ICs high parallelism is used to keep test costs down at both wafer probe and final test,” said Mark Kahwati, product marketing director for semiconductor testing group at Teradyne.
Final test can be performed at a range of temperatures. Some package defects can be easier to detect with a pin leakage test at cold temperature, due to the lower background current in CMOS devices. Yet the test budget may necessitate engineers finding a means to detect these at only one temperature.
Tracing root causes of failure
To reduce assembly defects causing automotive IC returns, engineers need to determine the root cause. This requires an understanding of failure mechanisms, in addition to knowing where the defect was generated. Connecting the defect to the point of generation in the various manufacturing steps data requires traceability.
First you need the data. For many decades, assembly engineers relied upon final test and inspection at the end of assembly as their sole data source. Data from the assembly equipment has been lagging.
“Assembly tools typically don’t have good communication interfaces. They just haven’t been very good at collecting data and supporting the analysis. But that’s changing, and thus resulting in a lot more data is now being collected. Especially from the more advanced packaging steps like die attach and wire bond, engineers now have access to time series, FDC data, and metrology data,” said Huntley. “I am chairing the SEMI Advanced Backend Factory Integration (ABFI) task force, which is focused on driving the level of automation and data collection that we have had in wafer fabs for years to now be done in assembly. In addition, the SEMI E142 standard has recently been enhanced so it also can be associated with an individual wire, which means the data per wire bond is inherently linked to the identity thread.”
Identity threads assign a name of a single die or unit across the supply chain. Names/identities change as units transverse the various manufacturing steps. Associating manufacturing data with the identity enables connection of the disparate data sources for engineering analysis.
“To do some more engineering on that particular part, you absolutely need an identity thread. Otherwise, you’re wasting your time. If you are unable to join the dots back to exactly, let’s say, a single wire in a single package, then you can’t possibly learn what was the root cause exactly, and what you can do to fix it (or detect it),” said Huntley. “But for the parts per billion, we’re going to have to learn from what happens out there in the field — and what actually is the root cause for this particular one-in-a-billion case.”
For assembly facilities to deliver 10 dppb quality to their automotive customers, they need to learn from customer returns. This requires investment in assembly equipment data collection and traceability. Latent defects that become activated during the warranty period yet pass electrical test necessitates 100% inspection to screen for these failures.
Yet all this investment in more inspection and data collection places a financial strain on traditionally inexpensive assembly operations. There is constructive tension between assembly facilities and their automotive customers, as they are both cost-sensitive. Still, somehow this pathway to 10 dppb will be funded.
Finding Defects In IC Packages
New inspection tools on the way, but there may still be holes in coverage.
Auto Chipmakers Dig Down To 10ppb
Driving to 10 defective parts-per-billion quality is all about finding, predicting nuanced behavior in ICs.
Part Average Tests For Auto ICs Not Good Enough
Advanced node chips and packages require additional inspection, analysis and time, all of which adds cost.
Case Study — 3D Wire Bond Inspection And Metrology
How a large automotive supplier managed 3D wire bond inspection and metrology.
Achieving Success In Automotive Leadframe Packages
Options and best approaches for meeting quality and reliability requirements.
Unknowns Driving Up The Cost Of Auto IC Reliability
More testing, metrology, data are required, but some chips will still fail.
Chips Good Enough To Bet Your Life On
Experts at the Table: Strategies for improving automotive semiconductors.
Predicting And Avoiding Failures In Automotive Chips
Experts at the Table: New approaches rely on more and better data, but it also requires sharing of data across the supply chain.
Automotive Outlook: 2022
Short-term IC supply-chain problems and long-term architectural and business changes top the list of what’s ahead.
Advanced Packaging For Automotive Chips
Cost, thermal dissipation, and overall reliability will likely determine which technologies get used where.