Analog behavior is difficult to compress into simple pass/fail decisions that could reduce redundant coverage.
Key Takeaways
Analog and mixed-signal test has reached an important turning point. While a new standard enables engineers to quantify test coverage of these essential devices for the first time, it’s still difficult to distinguish normal process variation from a soft failure.
The new IEEE Standard for Analog Defect Modeling and Coverage (IEEE 2427-2025, released in January) establishes a more consistent framework for determining that defects are detected by tests of analog and mixed-signal circuits, bringing greater discipline to an area where coverage has historically been difficult to define. At the same time, semiconductor companies are looking for ways to use analytics, historical data, and more adaptive test flows to reduce cost without increasing the number of defective or marginal devices that escape into the field.
“IC makers have a budget for test costs,” said Don Blair, business development manager at Advantest. “They test as much as they can up to their budget. After that, they have to find some way to cut test costs, including cutting test time, shortening or removing tests.”
An adaptive test works best when the relationship between an input and its expected result is understood well enough for a test to be shortened or removed without losing meaningful coverage. Analog and mixed-signal devices complicate matters because the result often falls within an acceptable range, and the difference between a merely functional device and a premium device may be buried within that range.
That tradeoff becomes more difficult when measurements don’t produce a single expected result. A data converter, such as an ADC or DAC, may pass a test for integral nonlinearity or differential nonlinearity and still perform better or worse than another passing device. The difference can determine whether the component may be sold into a premium application, or whether a customer is willing to accept a reduced test flow based on historical confidence in the design and process.
“It is difficult for customers to apply data analytics to mixed-signal test because many mixed-signal tests are non-deterministic and more qualitative,” Blair said. “With deterministic tests, there is one input and, for a good device, one output. With non-deterministic tests, there is one input and, for a good device, a range of outputs. Some are better than others, and this is where the “qualitative” aspect comes in.”
Digital test is not free of ambiguity, but it has benefited for decades from structural approaches — scan test, built-in self-test (BiST), stuck-at fault models, and clearly defined coverage metrics. Analog and mixed-signal devices still rely heavily on functional and specification-based measurements, particularly when the relevant behavior depends on gain, offset, leakage, noise, resistance, linearity, impedance, temperature, or operating conditions.
Adaptive test remains relevant to those devices, but the burden of proof is higher because the measurements being removed or redirected may be the only evidence that a marginal device will behave correctly in its intended application.
The economics of coverage
The cost calculation extends beyond the time required to run an individual test. Test equipment is a capital expense, but the full calculation also includes test-cell utilization, labor, power, floor space, and the cost of tying up expensive equipment with measurements that may no longer add useful information. Those costs scale differently depending on the device’s selling price and the end customer’s expected quality level.
A $10 device with a test-cost budget of 5% allows approximately $0.50 per device. A premium version selling for $20 yields approximately $1.00 at the same percentage, making additional screening economically feasible. The logic sounds simple until the test engineer has to decide which measurements are redundant and which protect the device margin that justifies the premium price.
The answer also changes as products mature. Early production flows often emphasize time-to-market, characterization, and confidence in the process, particularly for devices intended for automotive, industrial, and safety-critical applications. As the product stabilizes, manufacturers begin to look more closely at which measurements continue to provide enough value to justify their cost.
“It really depends on the market,” said Damian Megna, product manager for power and thermal solutions at Teradyne. “In automotive, it’s always in the spirit of increased coverage. But once industries mature, customers ask us to help determine what test coverage is redundant and can be eliminated as they return to the fundamental economics of cost of test and throughput.”
Coverage strategies vary widely between manufacturers and product families. Some customers perform extensive testing because the downstream cost of an escape is high; others rely on design margin, process stability, or internal analytics. The specific decisions are usually proprietary because they reveal where a manufacturer believes the real risks are located and which risks it is willing to manage statistically.
“Which tests they actually do and which ones may be guaranteed by history or data analytics is usually closely guarded information,” Blair said. “It depends on the desired quality level of the end product, and often this is determined by the end customer. Some demand higher levels of quality, and they want to see hard data for all of these mixed signal tests, while more cost-sensitive customers may not.”
The analog coverage problem
The difficulty begins with the meaning of coverage itself. Digital testing has long relied on structural techniques that make it possible to ask whether a defined set of faults was detected. Analog and mixed-signal devices do not always respond to that kind of accounting. Their failures may appear as shifts in a variety of areas, and the importance of any one shift depends on how the device will be used.
That makes it harder to determine whether two tests are redundant, whether a specification test can be shortened, or whether an upstream measurement can safely stand in for a later one. The problem is not simply that analog devices require more testing. The larger issue is that confidence has often been built through accumulated measurements because the industry has lacked a consistent way to determine which tests contribute meaningful defect coverage and which ones primarily add time.
“The greatest mismatch arises from the absence of objective, unbiased coverage metrics for A/MS circuits,” said Étienne Racine, product manager for Siemens EDA’s Tessent product line. “Without them, engineers often err on the side of caution, adding excessive tests that drive up both test times and overall costs and reduce yield.”
Fig. 1: Analog/mixed signal design. Source: Siemens EDA
IEEE 2427-2025 begins to address that gap by giving engineering teams a more disciplined way of asking whether a particular test is detecting a meaningful class of defects, whether that defect can be observed through another mechanism, and whether coverage can be improved earlier in the design flow.
The distinction matters because analog and mixed-signal tests do not fall into a single category. Some defects are best captured by direct specification-based measurements, while others can be detected by structural methods if the design provides sufficient observability and controllability. The most effective strategy is likely to combine structural, functional, and parametric tests. The underlying failure mechanism determines what can be moved, shortened, or inferred.
“Parametric defects, perhaps in differential transistor pairs or bandgap references, require specification tests, but they can be very quick DC tests,” Racine said. “Some aging-related defects require in-system parametric tests, but defects caused by electromigration or stress can cause shorts and opens detectable by in-system structural tests.”
That balance becomes more important as analog content is integrated into increasingly complex systems. A mixed-signal device may include digital control logic, power management, data conversion, sensing, RF functions, and safety features inside the same package. Each block may require a different balance between structural coverage, specification testing, and application-specific screening.
“Mixed-signal testing is almost always functional,” said Jack Lewis, director of applications and product management at Modus Test. “The chip itself is going to be functionally tested to its specs. It depends on whether it’s commercial grade or automotive, or if it’s safety-critical.”
This is where adaptive test becomes more complicated than simple test reduction. A manufacturer may be able to remove a redundant digital pattern after enough data has been collected, but an analog test may be the only measurement that reveals a marginal condition under a specific load, temperature, or voltage. The decision to reduce coverage, therefore, depends on whether the remaining measurements preserve the ability to distinguish normal process variation from a device that is drifting toward the edge of its usable specification.
When the measurement path becomes the variable
The problem becomes more difficult when uncertainty is introduced by the test path rather than the device. Every electrical measurement travels through a chain that includes the instrument, cables, probe card or socket, contacts, and the device pin itself. Each part of that chain can introduce resistance, contamination, wear, or coplanarity issues that affect the result.
That distinction is especially important in analog and mixed-signal test because the relevant margins may be small. A digital signal can often tolerate modest variation before the tester interprets the result incorrectly, but a precision analog measurement may shift enough to alter a binning decision or create a false failure. Once bad data enters an adaptive model, the system may treat contact degradation or voltage loss as evidence of a silicon problem.
Kelvin connections help isolate the device from the electrical path by using separate force and sense connections. The force line applies the voltage or current, while the sense line confirms what actually reaches the device pin. Without that visibility, a tester may force the expected voltage upstream, while the device receives slightly less due to interconnect losses.
“In general, the best sockets you can have in a mixed-signal or analog space are Kelvin sockets, which have a force and a sense connection for every pin, ideally,” Lewis said. “The voltage accuracy requirements are very high, so that precision matters. The less precise you are, the more you have to guard-band.”
Guard-banding protects against uncertainty, but it also reduces the usable performance window. A device that could meet a customer’s specification may be rejected or sold into a lower-value bin because the test flow cannot distinguish device margin from measurement-path error. That tradeoff becomes more consequential as manufacturers use analytics to make finer-grained decisions about which devices need additional testing.
One challenge in mixed-signal and power devices is measuring RDSon — the drain-to-source resistance of the transistor during operation — which is often the most critical spec. Measurements run in the milliohm range, where any contact interruptions can corrupt the results.
“RDSon, for all these mixed-signal and power devices, is probably the most common interconnect-driven issue,” Lewis said. “Sometimes the package they’ve chosen doesn’t have enough pins to have a sense pin. So now they’ve got to deal with the Kelvin contactor — or they don’t want to spend money on Kelvin contactors because they’re much more expensive.”
High-pin-count digital devices create their own interconnect problems, but the analog challenge is of a different character.
“With high-pin-count digital, the challenge is isolation, coverage, and throughput with so many pins,” said Vidya Vijay, director of business development at Nordson Test and Inspection. “Analog and mixed-signal is a different problem entirely. Fewer pins, but zero margin for error. A single contact sitting outside its 0.25mm true position tolerance or breaching the contact height spec will corrupt the measurement completely, and there is nothing in the test architecture to absorb it.”
The need to isolate tester behavior from device behavior is therefore fundamental to any adaptive flow. As measurements become more precise and decisions become more automated, the quality of that distinction will determine whether adaptive test improves yield or merely shifts uncertainty into a more complicated part of the process.
“Customers are really pressing us for proof in terms of characterization and specification, to build confidence that the instrument itself is repeatable under all the required environmental conditions,” said Teradyne’s Megna. “They want to know that anything they see that’s an anomaly is related to their device, not the tester.”
Correlating physical variation with electrical behavior
Adaptive test depends on the strength of the correlations behind each decision. A test can be shortened, redirected, or removed only when earlier process, metrology, or electrical data can predict the later result with enough confidence. That becomes harder for analog and mixed-signal devices because subtle material variation, package stress, and buried defects may shift electrical behavior without producing an obvious physical failure.
The challenge grows as devices incorporate thinner films, denser interconnects, and more heterogeneous material stacks. A small variation introduced early in the process may not become visible until later electrical test, thermal loading, or field operation. By then, the original cause may be buried beneath additional layers or distributed across multiple process steps.
“The novel architectures and materials in the 3D-device era could introduce additional potential ‘escape tunnels’ that cause unattributable performance deterioration or yield loss,” said Lei Zhong, product marketing senior director at Onto Innovation. “These escape tunnels can make it more difficult to correlate a specific process excursion with the electrical behavior that appears later in the flow.”
Failure analysis can help close that gap by connecting downstream electrical behavior to repeatable upstream signatures. Once that relationship is understood, the physical or chemical signature can be incorporated into the screening strategy, allowing test engineers to distinguish meaningful variation from background process noise.
“When you do this rear-view-mirror approach, you can then say, ‘This bonding structure, this molecular environment, corresponds to positive electrical performance downstream; versus this process does not correspond to positive performance,’” said Cassandra Phillips, product manager for nano IR systems at Bruker. “When you initially have that question answered through the failure-analysis approach, you can then implement that as screening criteria.”
That distinction matters because analog and mixed-signal devices often fail gradually rather than catastrophically. Engineers need to understand how far a device has moved from its expected behavior, whether that deviation is stable, and whether it is likely to worsen under operating conditions. A small shift in resistance, leakage, gain, or noise may be as important as an obvious open or short.
“With a digital chip, you’ve pretty much got open lines and shorts,” said Thomas Rodgers, senior director of market strategy and head of business sector electronics at ZEISS Microscopy. “With an analog device, you’re also really interested in deviations from an expected resistance. That’s a much more difficult thing to characterize and measure.”
The value of metrology and failure analysis, therefore, extends beyond root-cause investigation. When a physical signature can be tied reliably to an electrical outcome, it can become part of the evidence used to focus test time where the risk is highest. When that correlation remains weak, direct measurement is still the safer approach.
Production test cannot capture every use condition
Production test also captures only a limited view of how complex devices will behave over time. Embedded monitoring is increasingly being used in advanced SoCs to track timing margin, voltage, thermal conditions, workload stress, and aging during mission-mode operation. Although these monitors do not replace direct analog and mixed-signal measurements, they illustrate the growing value of extending visibility beyond a single production insertion.
Embedded telemetry and lifecycle monitoring can extend the visibility of the test flow beyond the factory. They do not replace production test, but they can provide additional context for understanding how voltage, temperature, noise, and workload affect long-term behavior and eventually improve screening rules by showing which upstream signatures were genuinely predictive. Thermal control remains a critical part of that equation because temperature affects both the measurement and the behavior of the device being measured.
“They want much better, more accurate thermal control than they have today, because they don’t want large guard bands,” Megna said. “It helps their yield, but it also helps reliability. If they can accurately control the temperature, and they can reliably get a test insertion that’s truly at hot, then if that device passes, they’ve provided better future-proofing for reliability.”
The relationship between yield and reliability is especially important for analog and mixed-signal devices. Tighter thermal control may reduce false failures during production, but it also provides a clearer picture of whether the device margin is real. The same accuracy that protects yield can improve confidence that a passing device will remain inside its specification after it leaves the factory.
CPO concentrates the same problems
Co-packaged optics provides a useful example of where all these challenges are heading. It brings multiple difficult variables together inside the same package. Electrical behavior, optical performance, alignment, thermal sensitivity, signal integrity, power integrity, assembly variation, and package stress must all be understood well enough that test decisions remain meaningful. The addition of photonics makes existing multi-physics challenges more acute.
“Photonics is very sensitive to heat, so thermal becomes even more important,” said Amlendu Shekhar Choubey, senior director of product management at Synopsys. “You need to have an integrated flow where optical simulation and electrical simulation can coexist, and then a design platform where you can integrate electronic design, advanced packaging, and PIC design so that you can co-design all these components from architecture through final signoff.”
Fig. 2: Analog IP diagram showing open “eye” waveforms, which represent signal quality and integrity. Source: Synopsys
CPO does not create an entirely separate test problem. It intensifies the same correlation problem already present in analog and mixed-signal manufacturing by adding optical performance to a system where electrical, thermal, and package-level variations are already difficult to parse.
“It is not something which is totally different,” Choubey said. “It just makes some of the challenges more acute.”
Building confidence earlier
Analog and mixed-signal devices are unlikely to follow a single path toward adaptive test. Some products will continue to require extensive functional testing because the application leaves little room for uncertainty. Others will use structural methods, improved modeling, upstream metrology, and better data correlation to reduce redundant measurements or direct test time toward the devices carrying the greatest risk.
The most productive shift will begin earlier in the design flow. Better analog defect models, more consistent coverage accounting, and improved observability can help engineering teams identify which measurements are essential before the device reaches production. That does not eliminate the need for specification-based testing, but it gives test engineers a clearer basis for deciding where direct measurement is necessary and where another source of evidence may be sufficient.
“Adopting a shift-left approach, where test coverage is measured before design verification, before layout, or at least before tape-out, enables teams to address coverage gaps proactively, rather than waiting until silicon is ready,” said Racine.
The pressure to improve throughput will continue because test cost remains part of the economics of every device. Analog and mixed-signal manufacturers have strong reasons to use analytics, historical data, and more selective test flows, particularly as packages become more complex and new applications demand tighter margins. The difficulty lies in proving that a measurement is redundant before removing it, separating device behavior from test-path variation, and understanding which upstream signals are predictive enough to support a decision.
Analog and mixed-signal devices do not sit outside the evolution toward smarter test. They expose the places where that evolution still needs work.
Related Articles
Transforming Test For Co-Packaged Optics
Profound changes are underway to ensure the reliability of co-packaged opto-electronic systems
Smart Test Collides With The Data Chain
Increasing complexity is limiting the ability of machine learning models to effectively utilize test data.
Promises And Perils Of Parallel Test
Test costs may be reduced, but how much depends on a whole bunch of factors.
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Leave a Reply