Electronics Guide

Articles in This Category

Traceability and the Realization of Units

Traceability is a chain, and the metaphor is worth taking literally. A production tester on a factory floor is calibrated against a laboratory reference standard, which is calibrated by an accredited calibration laboratory, whose own standards are calibrated by or compared against a national metrology institute, which realizes the unit directly. Each link is a calibration, each calibration is documented, and each contributes uncertainty. Uncertainty therefore grows monotonically from the top of the chain to the bottom, which is why a bench multimeter cannot be more accurate than the standard that calibrated it and why claims of traceability that omit an uncertainty figure convey no information at all.

What sits at the top of the chain changed on 20 May 2019, when the revised SI entered into force. All seven base units are now defined by fixing the numerical values of seven defining constants: the caesium-133 hyperfine transition frequency for the second, the speed of light for the metre, the Planck constant for the kilogram, the elementary charge for the ampere, the Boltzmann constant for the kelvin, the Avogadro constant for the mole, and the luminous efficacy of 540 terahertz radiation for the candela. The practical effect is that units are no longer tied to artifacts. The international prototype kilogram, a platinum-iridium cylinder held near Paris, ceased to define mass; any laboratory with sufficient capability can now realize the unit from physical constants rather than by comparison against a single object.

Electrical metrology benefited most directly, because it had already been running on quantum realizations for decades. The alternating-current Josephson effect relates voltage to frequency through the Josephson constant, and the quantum Hall effect relates resistance to fundamental constants through the von Klitzing constant. Both constants are combinations of the Planck constant and the elementary charge, so fixing those two values in 2019 made the Josephson and quantum Hall realizations primary rather than conventional. Before that date, laboratories worked against agreed conventional values adopted in 1990, which produced small but real offsets between the practical electrical units and the SI. Those offsets are now gone. Most laboratories realize the ampere indirectly, from a Josephson voltage standard and a quantum Hall resistance standard through Ohm's law, rather than by counting electrons. The underlying physics is treated under precision measurement, and the instruments that carry these units into working laboratories under calibration and standards equipment.

Time and frequency deserve separate mention, because they are by a wide margin the best-realized quantities in the SI and because so many other measurements are converted into them. Caesium fountain primary standards and, increasingly, optical clocks reach fractional uncertainties far below anything achievable in voltage or resistance. Designers exploit this whenever they can: a quantity measured as a frequency, a period, or a phase inherits an accuracy that the same quantity measured as an amplitude cannot approach. That asymmetry explains the architecture of many precision instruments, as described under time and frequency measurement.

Uncertainty, Decision Rules, and Guardbanding

The reference document for uncertainty evaluation is the Guide to the Expression of Uncertainty in Measurement, issued by the Joint Committee for Guides in Metrology as JCGM 100:2008 and republished as ISO/IEC Guide 98-3. Its central move is to treat every contribution to doubt in the same currency, the standard uncertainty, regardless of where the doubt came from. Type A evaluation derives a standard uncertainty statistically, from repeated observations. Type B evaluation derives one from any other source: a calibration certificate, a manufacturer's specification, a handbook value, the resolution of a display, or reasoned judgment. The distinction concerns the method of evaluation, not the quality or the nature of the component, and it is routinely misread as a distinction between random and systematic effects.

An uncertainty budget enumerates the contributions, assigns each a distribution, converts each to a standard uncertainty, and combines them. Typical entries for an electrical calibration include the uncertainty of the reference standard as stated on its certificate, the drift of that standard since it was calibrated, the resolution and repeatability of the instrument under test, thermal electromotive forces in connections, loading and lead resistance effects, cable and connector repeatability at radio frequencies, and the sensitivity of the whole arrangement to temperature and humidity. A certificate value stated at a coverage factor of two is divided by two to recover the standard uncertainty; a resolution limit or a specification band with no other information is normally treated as a rectangular distribution, whose half-width is divided by the square root of three. The components combine in quadrature through the law of propagation of uncertainty, and the result is multiplied by a coverage factor, conventionally two, to give an expanded uncertainty covering roughly ninety-five percent of the distribution when that distribution is approximately normal.

The linearized propagation at the core of the method has known limits. Where the measurement model is strongly nonlinear, or where the dominant contribution is far from normal, JCGM 101:2008 supplies a Monte Carlo method that propagates distributions numerically instead of propagating first-order sensitivity coefficients. The guides remain under active maintenance: the International Committee for Weights and Measures approved an amendment to JCGM 100:2008 addressing the treatment of nonlinear models, and JCGM 101 is undergoing revision. Practitioners should confirm the current issue rather than assume the 2008 text is the last word.

Uncertainty becomes consequential the moment a laboratory states that something passes. ISO/IEC 17025:2017 requires that when a laboratory issues a statement of conformity to a specification, it document the decision rule it applied and take account of the level of risk that rule accepts. ILAC-G8 provides the standard guidance. Simple acceptance, sometimes called shared risk, compares the measured value with the tolerance limit and ignores uncertainty entirely; it is defensible only when the uncertainty is small relative to the tolerance and the customer has agreed. Guardbanding pulls the acceptance limits inward by some fraction of the expanded uncertainty, trading a higher rate of false rejections for a lower rate of false acceptances.

Two quantities govern the trade. The test uncertainty ratio compares the tolerance of the item under test with the uncertainty of the measurement, and a ratio of four to one is the traditional rule of thumb for treating uncertainty as negligible. The probability of false accept, sometimes called consumer risk, states directly what fraction of accepted items are actually out of tolerance. ANSI/NCSL Z540.3-2006, the American national standard for calibration of measuring and test equipment, requires that the probability of false accept not exceed two percent and be documented, and permits a test uncertainty ratio of at least four to one as the fallback where estimating that probability is not practicable. The choice matters commercially as well as technically, because tighter guardbands push instruments into repair or adjustment that a looser rule would have passed.

Accreditation and Mutual Recognition

ISO/IEC 17025:2017 is the third edition of the standard and restructured it substantially. Its requirements fall into general clauses covering impartiality and confidentiality, structural requirements covering the laboratory's legal identity and organization, resource requirements covering personnel, facilities, equipment, metrological traceability, and externally provided products and services, process requirements covering everything from contract review through method validation, sampling, handling of items, uncertainty evaluation, assurance of the validity of results, reporting, complaints, and nonconforming work, and management system requirements offered in two forms. Option A prescribes the management system elements directly; Option B allows a laboratory that already operates a management system conforming to ISO 9001 to satisfy the clause through it. The 2017 edition also replaced the earlier prescriptive treatment of preventive action with risk-based thinking, obliging the laboratory to identify risks and opportunities relevant to its activities rather than to run a separately named procedure.

Accreditation and certification are frequently confused, and the difference is practical. Certification attests that a product, process, or management system conforms to a specified standard. Accreditation attests to technical competence for a defined scope of activity, and the scope document is the operative one. A laboratory accredited for direct-current voltage to a stated range and uncertainty is not thereby accredited for radio-frequency power, and a certificate that falls outside the accredited scope carries no accreditation status even though the same laboratory issued it. Buyers of calibration services who check only for a logo, and never read the scope, routinely obtain less than they believe they have bought. The mechanics of accreditation and certification bodies are covered under certification body processes.

Two international arrangements make the results portable. The International Laboratory Accreditation Cooperation operates a mutual recognition arrangement under which signatory accreditation bodies, now numbering more than one hundred across a comparable number of economies, evaluate one another by peer assessment and accept one another's accredited reports and certificates. Above it, the arrangement drawn up by the International Committee for Weights and Measures in 1999 covers the national metrology institutes themselves: their calibration and measurement capabilities are peer-reviewed, supported by the results of key comparisons, and published in a public database. Together these arrangements are what allow a calibration performed in one country to be accepted in another without repetition, the same principle that governs test reports under mutual recognition agreements. In the United States, accreditation is offered by the National Voluntary Laboratory Accreditation Program operated by NIST under NIST Handbook 150, and by private bodies including A2LA and the ANSI National Accreditation Board.

Accreditation is periodic; competence must be demonstrated continuously. Proficiency testing and inter-laboratory comparison supply that ongoing evidence by sending an artifact of known or consensus value around a group of laboratories and scoring the results. ISO/IEC 17043 governs the providers of such schemes and ISO 13528 the statistical methods used to evaluate them. Testing laboratories are commonly scored with a z-score, where a magnitude at or below two is treated as satisfactory, between two and three as questionable, and at or above three as unsatisfactory. Calibration laboratories more often use the normalized error, which divides the difference between the participant's result and the reference value by the root sum of squares of the two expanded uncertainties; a magnitude at or below one is satisfactory. The normalized error is the more searching test, because a laboratory can fail it not by measuring badly but by understating its own uncertainty.

Sustaining the Measurement System

A calibration is a statement about the past. It records how an instrument behaved on the day it was measured, and its value decays as the instrument drifts, is handled, is transported, or is overloaded. Everything in this section exists to manage that decay.

Calibration intervals are the first control. An initial interval is normally taken from the manufacturer's specification, from the interval over which the manufacturer states the accuracy, or from established practice for the instrument class. It should not remain there. NCSL International Recommended Practice RP-1 catalogs the methods for adjusting intervals from evidence, ranging from simple reactive schemes that lengthen the interval after a run of in-tolerance returns and shorten it after a failure, to statistical approaches that fit an observed reliability model to the accumulated as-found data and solve for the interval that meets a chosen in-tolerance probability. All of them depend on one practice: recording the as-found reading before any adjustment, alongside the as-left reading afterward. A laboratory that adjusts first and records only the as-left condition has destroyed the only data that could have justified its intervals.

As-found data serves a second and more urgent purpose. When an instrument returns significantly out of tolerance, every measurement made with it since its previous calibration is suspect. ISO/IEC 17025 obliges the laboratory to evaluate that nonconforming work, to assess its significance, and where necessary to notify the customers whose results are affected. Doing so requires knowing which measurements the instrument touched, which in turn requires records that link instruments to jobs. This reverse traceability, sometimes called equipment recall, is the reason quality systems insist on identifying the specific asset used for each measurement rather than merely the instrument model, and it connects directly to the corrective action machinery described under failure reporting and corrective action.

Between calibrations, intermediate checks bound the exposure. A stable check standard measured on a fixed schedule and plotted on a control chart will reveal drift, a damaged connector, or a shifted range long before the next scheduled calibration, and it converts the annual calibration from a leap of faith into a confirmation. Environmental control serves the same end. Temperature, humidity, vibration, electromagnetic interference, and the quality of the supply all shift readings, and stability generally matters more than the absolute set point: a laboratory held steadily at a temperature slightly away from its nominal value will outperform one that wanders through the nominal value twice a day. ISO 1 fixes twenty degrees Celsius as the standard reference temperature for dimensional work, while electrical laboratories commonly specify a band around twenty-three degrees Celsius, and calibration certificates state the conditions under which the measurement was made precisely because the values do not travel unchanged to other conditions.

On the production side, the equivalent discipline is measurement system analysis. A gauge repeatability and reproducibility study, in the form described in the automotive industry's measurement systems analysis reference manual, partitions observed variation into repeatability, the variation of the same operator measuring the same part repeatedly, and reproducibility, the variation between operators. Common acceptance guidance treats a combined figure below ten percent of the tolerance or of the total study variation as acceptable, ten to thirty percent as conditionally acceptable depending on the criticality and the cost of improvement, and above thirty percent as unacceptable. The number of distinct categories the system can resolve is reported alongside it. The point of the study is uncomfortable but essential: a process that appears to be producing scattered parts is sometimes producing consistent parts and measuring them badly, and no amount of process adjustment will fix that.

From Measurement to Investigation

Failure investigation inverts the discipline. Instead of proving that a number can be trusted, the investigator must prove that a conclusion can be trusted, and the constraint is that evidence is consumed by examination. Every step changes the item: cleaning removes residues that might have identified a contaminant, powering the unit can erase the signature of the fault or destroy the remaining evidence of it, decapsulation exposes a die but eliminates the package, and cross-sectioning is irreversible. The governing rule follows directly. Work from least invasive to most, and do not take a step until the information available from the previous step has been captured.

A typical sequence for an electronic assembly begins with receipt and documentation: photographing the item as received, recording packaging and labeling, observing electrostatic discharge precautions, and noting anything the return path may have altered. Non-destructive electrical characterization follows, including curve tracing of accessible pins and functional testing where doing so will not propagate damage. External visual and stereomicroscopic examination comes next, then radiography and computed tomography to see internal structure, solder voiding, and cracked interconnects without opening anything, and scanning acoustic microscopy to reveal delamination and voids within packages. Fault isolation techniques such as lock-in thermography and photon emission microscopy localize a defect to a small area while the device remains intact. Only then does the work turn destructive: decapsulation, cross-sectioning, focused ion beam milling, and scanning electron microscopy with energy-dispersive X-ray spectroscopy for elemental identification. The instruments involved are surveyed under inspection and analysis equipment, and the techniques in depth under physical failure analysis and semiconductor failure analysis.

Fractography and metallurgical examination read the surfaces that the failure created. Fatigue leaves progression marks visible to the eye and striations visible under an electron microscope, with an origin that can usually be traced back to a stress concentration or a defect. Ductile overload leaves dimpled rupture. Brittle fracture leaves cleavage facets or, where the path follows grain boundaries, an intergranular surface that points toward embrittlement or stress corrosion. In electronics assemblies the same logic applies at smaller scale to solder joints, where thermal cycling produces characteristic grain coarsening and cracks that propagate near the intermetallic layer, and where the fracture path distinguishes a joint that failed from fatigue from one that was never properly formed. That subject is treated under solder joint and interconnect analysis.

Custody runs alongside the technical work and is often the part that decides whether the technical work survives challenge. A chain of custody is an unbroken documented record of who held the evidence, when, and for what purpose, with each transfer signed and dated and storage controlled against tampering, contamination, and loss. Where litigation is possible, the obligation extends further. Parties are expected to preserve relevant evidence once a dispute is reasonably anticipated, and destroying or altering it exposes them to sanctions for spoliation. In practice this means that a destructive examination of a disputed item is arranged with notice to the other parties, is frequently conducted with their representatives present or their protocol agreed in advance, and is documented at every stage.

Making Findings Defensible

An investigation ends in a written conclusion, and the standards that govern that conclusion are as important as the ones that governed the laboratory work. Several ASTM standard practices address the sequence directly, including E1188 on the collection and preservation of information and physical items by a technical investigator, E860 on examining and preparing items that are or may become involved in litigation, E620 on reporting the opinions of scientific or technical experts, E2713 as a general guide to forensic engineering, and E3176 on forensic engineering expert reports. Practitioners should confirm current status before citing any of them, since the portfolio is actively maintained and two long-standing practices, E678 on the evaluation of scientific or technical data and E1020 on reporting incidents that may involve litigation, were withdrawn in 2022 with their guidance absorbed elsewhere.

In United States federal courts, admissibility of expert opinion is governed by Federal Rule of Evidence 702, as interpreted by the Supreme Court in Daubert v. Merrell Dow Pharmaceuticals and the decisions that followed it in General Electric Co. v. Joiner and Kumho Tire Co. v. Carmichael. The rule was amended effective 1 December 2023 to state explicitly that the proponent must demonstrate to the court, by a preponderance of the evidence, that each admissibility requirement is met, and that the expert's opinion must reflect a reliable application of the principles and methods to the facts of the case. The amendment did not change the underlying standard; it corrected a pattern of decisions that had treated expert testimony as presumptively admissible and had deferred reliability questions to cross-examination. Some state courts continue to apply the older general-acceptance test derived from Frye v. United States. The engineer's practical takeaway is unchanged by the jurisdictional detail: use methods with established acceptance and known limitations, document the examination sequence, separate observation from inference in the report, state the alternative hypotheses considered and the evidence that excluded them, and decline to opine beyond what the data support. Expert-witness practice and the litigation posture that surrounds it are covered under product liability prevention and forensic engineering and investigation.

One conceptual distinction underpins the entire reporting exercise. A failure mode is what the item did: the joint cracked, the die shorted, the electrolytic capacitor vented. A failure mechanism is the physical process that produced it: thermomechanical fatigue, electromigration, electrolyte dry-out. A root cause is the reason the mechanism was allowed to operate: an unqualified thermal profile, a design margin consumed by a tolerance stack, a supplier change that passed without requalification. Corrective action attaches only to the last of the three, and an investigation that stops at the failure mode produces a report that satisfies curiosity while leaving the population unchanged. Structured methods including the eight disciplines process, cause-and-effect diagrams, and fault tree analysis exist to force the analysis past that stopping point; the five-whys technique is the most popular and the weakest, because it follows a single causal path and terminates at whatever answer sounds satisfying. These methods are compared under root cause analysis techniques.

Where the Discipline Fails

The failure patterns in this field are consistent enough to list. Traceability is treated as paperwork, satisfied by a certificate filed in a binder that nobody has read, whose uncertainty statement nobody has compared against the tolerance it is supposed to support, and whose accredited scope nobody has checked against the parameter actually measured. Calibration is performed to the manufacturer's specification without confirming that the specification covers the range, frequency, or loading condition in use. Drift between calibrations is ignored because no check standard exists to reveal it. Conformity is declared without a documented decision rule, leaving the consumer risk unquantified and, when a dispute arises, indefensible.

Investigation has its own catalog. Destructive work is performed before non-destructive work, which is the one error in failure analysis that cannot be undone. A single unit is analyzed and its findings generalized to a population that was never sampled. The failed part is replaced, the symptom disappears, and the disappearance is recorded as confirmation of a cause that was never actually tested. Correlation is presented as causation because a supplier change and a failure rate increase happened in the same quarter. Underlying most of these is a governance problem rather than a technical one: investigations commissioned by the party whose work is under examination, and calibration intervals set by whoever bears the cost of shortening them.

About This Category

Specialized testing and analysis represents the technical foundation upon which all measurement activity rests. Without sound calibration, documented traceability, and adherence to metrology standards, test results lack the credibility that regulatory submissions, quality decisions, and engineering judgments require. Equally, when products fail, disciplined forensic investigation turns ambiguous symptoms into actionable root causes. The two subcategories are halves of one practice: calibration and metrology standards establish that a number means what it claims to mean, and forensic and failure investigation standards establish that a conclusion drawn from numbers will withstand examination by a regulator, an auditor, or a court.

Related material appears under testing and certification, which produces the compliance evidence that this category underwrites, quality management systems, which house the records and the corrective action machinery, international standards organizations, which publish the standards themselves, and failure analysis methodologies, which treat the investigative techniques in technical detail. This category addresses the often-overlooked infrastructure of measurement science and failure analysis that makes reliable testing possible, knowledge that is essential for anyone responsible for laboratory operations, quality management, or regulatory compliance in the electronics industry.