15. Reliability, Engineering Decisions and Emerging Electronics

Introduction and Context

Electronic-device engineering has traditionally emphasized electrical performance: gain, speed, power consumption, noise, breakdown voltage, and cost. In industrial systems, however, a device that meets its data-sheet specifications but fails prematurely is not a successful engineering product. Reliability, manufacturability, supply continuity, lifecycle cost, regulatory compliance, environmental impact, and ethical deployment are equally important design dimensions.

This chapter considers electronics as a lifecycle system rather than an isolated circuit. A transistor, sensor, power module, or integrated circuit passes through design, simulation, fabrication, assembly, qualification, field operation, maintenance, and end-of-life treatment. Decisions made in one phase influence risk in all later phases. For example, a narrow process margin may improve nominal performance but increase production variability; an inexpensive package may raise thermal resistance; dependence on a single overseas supplier may create strategic vulnerability even when the component is technically excellent.

The chapter also examines artificial-intelligence-assisted electronic design automation (EDA), digital twins, predictive maintenance, flexible and neuromorphic electronics, chiplet-based systems, quantum devices, and sustainable electronics. These technologies offer substantial opportunities, but they introduce new forms of uncertainty. Sound engineering therefore requires evidence, traceability, independent verification, and responsible governance rather than uncritical adoption of novelty.

Learning Objectives

After studying this chapter, the student should be able to:

1. Explain reliability terminology, failure mechanisms, and reliability models used for electronic devices and systems.

2. Distinguish reliability, availability, maintainability, quality, and safety.

3. Evaluate engineering alternatives using lifecycle cost, manufacturability, supply-chain risk, and technical performance.

4. Apply qualification, accelerated testing, statistical analysis, and root-cause methods.

5. Describe responsible uses and limitations of AI-assisted EDA, digital twins, and predictive analytics.

6. Compare emerging technologies including flexible electronics, neuromorphic devices, chiplets, and quantum devices.

7. Incorporate sustainability, ethics, standards, cybersecurity, and governance into electronics decisions.

Reliability as a Design Property

Reliability is the probability that an item performs its intended function for a specified period, under stated conditions. It is therefore not an intrinsic label such as “good” or “bad”; it is conditional on stress, environment, duty cycle, manufacturing quality, and the definition of failure. A power converter may be reliable at its rated temperature but unreliable in an enclosure with inadequate cooling.

If failures are statistically described by a constant failure rate λ, the reliability function is

R(t) = exp(−λt),

and the mean time to failure is approximately 1/λ. The constant-rate assumption is useful for random failures but does not describe all products. The familiar bathtub curve contains an early-life region dominated by manufacturing defects, a useful-life region with approximately random failures, and a wear-out region in which ageing mechanisms such as electromigration, dielectric breakdown, fatigue, and corrosion become increasingly important.

For repairable systems, availability is often more useful than reliability:

A = MTBF/(MTBF + MTTR),

where MTBF is mean time between failures and MTTR is mean time to repair or restore service. A highly reliable but difficult-to-repair system may have lower operational availability than a moderately reliable modular system that can be rapidly serviced. Maintainability, diagnostics, redundancy, graceful degradation, and spare-parts planning therefore belong in the architecture discussion.

Quality and reliability must also be separated. Quality concerns conformity at manufacture; reliability concerns continued performance over time. A product may have excellent initial quality but poor field reliability because its materials age incompatibly. Conversely, screening and burn-in may remove early failures without eliminating a fundamental wear-out mechanism.

Failure Mechanisms and Physics of Degradation

Failure analysis begins by identifying the physical mechanism, the applied stress, and the vulnerable structure. Common semiconductor mechanisms include electromigration in metal interconnects, time-dependent dielectric breakdown, bias-temperature instability, hot-carrier degradation, junction leakage, latch-up, and electrostatic-discharge damage. Their rates depend on current density, electric field, temperature, voltage, geometry, and process conditions.

Packaging frequently determines system reliability. Repeated thermal expansion creates solder-joint fatigue; wire bonds may lift or break; mould compounds can absorb moisture; interfaces can delaminate; and corrosion may occur when contamination, humidity, and electrical bias coexist. In power electronics, thermal cycling and high temperature are particularly significant because losses generate self-heating and temperature gradients. A design with lower electrical loss can still fail if its thermal path is poorly engineered.

Printed circuit boards experience vibration, shock, connector fretting, ionic contamination, and solder fatigue. Sensors may drift because of material ageing, mechanical stress, radiation, or calibration instability. Displays and flexible electronics add bending, stretching, barrier-layer cracking, and repeated mechanical cycling to the reliability problem.

A useful engineering practice is to construct a stress–mechanism–failure chain. For example:

• Stress: high junction temperature.
• Mechanism: accelerated diffusion and interconnect degradation.
• Observable symptom: increased resistance and leakage.
• Functional failure: timing error or thermal shutdown.
• System consequence: intermittent reset or loss of control.

Arrhenius acceleration is often used for thermally activated mechanisms:

AF = exp[(Ea/k)(1/Tuse − 1/Ttest)],

where Ea is activation energy, k is Boltzmann’s constant, and temperatures are in kelvin. The equation is useful only when the same failure mechanism operates at both temperatures. Excessive test temperature can introduce a different mechanism and make extrapolation misleading. Voltage, humidity, current, and mechanical stresses may require other models or combined-stress experiments.

Reliability Engineering Methods

A reliability programme begins during requirements definition. The engineer should specify mission profile, environmental limits, duty cycle, acceptable failure rate, service life, maintenance concept, and safety consequences. Reliability allocation then distributes system targets among subsystems according to architecture and criticality.

Failure Mode and Effects Analysis (FMEA) is a bottom-up method. Each component or function is examined for possible failure modes, causes, local effects, system effects, detection methods, and controls. A risk priority number may be used as a screening aid, but it should not replace engineering judgement; multiplication of ordinal rankings can conceal important differences. Failure Modes, Effects, and Criticality Analysis gives greater attention to severity and mission consequence.

Fault-tree analysis is top-down. Starting with an undesirable event, such as loss of braking control or converter shutdown, the analyst decomposes logical relationships using AND and OR gates. It is especially useful for identifying combinations of failures and evaluating redundancy. Event-tree analysis proceeds from an initiating event toward possible consequences and protective responses.

Design for reliability includes derating, thermal management, robust margins, protection circuits, controlled interfaces, redundancy, error detection, and testability. Derating limits electrical, thermal, mechanical, or environmental stress below the component maximum. It improves margin but may increase size or cost. Redundancy can improve availability, yet common-cause failures—such as a shared power supply, software defect, or thermal environment—can defeat apparently independent channels.

Qualification testing demonstrates that a design and its manufacturing process can withstand specified stresses. Typical tests include temperature cycling, high-temperature operating life, humidity-bias exposure, mechanical shock, vibration, ESD, latch-up, power cycling, and solder-joint evaluation. Qualification is not a guarantee of field life. It is evidence whose validity depends on representative samples, realistic mission profiles, adequate sample size, and correct failure interpretation.

Standards provide shared methods and acceptance criteria. Depending on application, engineers may consult IEC, JEDEC, IPC, ISO, automotive, aerospace, medical-device, functional-safety, and electromagnetic-compatibility standards. A standard should be treated as a minimum framework, not a substitute for product-specific risk analysis. Compliance claims must identify the exact edition, test conditions, product configuration, and exclusions.

Engineering Decisions: Beyond the Purchase Price

A technically superior component is not automatically the best engineering choice. Multi-criteria decision analysis makes trade-offs explicit. Criteria may include performance, reliability, lifecycle cost, manufacturability, energy use, supply risk, repairability, cybersecurity, environmental impact, and regulatory suitability. Weighting should be agreed by stakeholders before alternatives are scored, and sensitivity analysis should determine whether the decision changes when assumptions vary.

Lifecycle cost includes acquisition, tooling, qualification, production, energy, maintenance, downtime, inventory, obsolescence, disposal, and the cost of failures experienced by customers. A lower unit price can be economically inferior if it requires expensive screening, causes more field returns, or has a short supply horizon. Reliability economics often uses expected cost:

Expected cost = acquisition cost + operating cost + probability of failure × consequence cost.

Supply-chain risk includes single-source dependence, geographic concentration, counterfeit exposure, long lead times, export controls, material scarcity, and supplier financial instability. Mitigation may involve second sourcing, approved alternates, buffer inventory, redesign for substitution, supplier audits, traceability, and contractual continuity provisions. These measures also have costs; resilience is not free and should be treated as an engineered capability.

Manufacturability connects design intent with production reality. Design for Manufacturing and Assembly seeks accessible test points, tolerant dimensions, standard packages, reduced part count, controlled process windows, and assembly methods compatible with available equipment. Design for Test improves fault observability and reduces diagnostic time. A design that cannot be economically inspected or repaired may have poor practical reliability even if laboratory performance is excellent.

AI-Assisted EDA, Digital Twins, and Predictive Analytics

Machine learning is increasingly used in circuit sizing, placement and routing, verification, yield prediction, defect classification, thermal optimization, and failure diagnosis. AI-assisted EDA can search large design spaces faster than manual iteration and can identify correlations that are difficult to detect using conventional heuristics. However, an optimization objective may produce undesirable results if it omits reliability, explainability, manufacturability, or worst-case constraints.

A digital twin is a computational representation connected, directly or indirectly, to a physical asset through design data and operational measurements. In electronics it may combine circuit models, thermal models, manufacturing history, sensor data, and maintenance records. A useful twin supports condition monitoring, remaining-useful-life estimation, design validation, or process optimization. It is not simply a three-dimensional model or a dashboard.

Predictive analytics estimates future condition from measured indicators such as temperature, current, vibration, error counts, leakage, or timing margin. Its value depends on data quality and failure labels. Rare failures create class imbalance; changing operating conditions create distribution shift; sensor drift can be mistaken for component degradation; and correlations can be confused with causes. Models must be validated on independent data and monitored after deployment.

Responsible AI-assisted engineering requires:

• traceable datasets, assumptions, model versions, and design changes;
• physical constraints and conservative limits built into optimization;
• independent verification against simulation, measurement, and expert review;
• uncertainty estimates rather than unsupported point predictions;
• protection of proprietary design and operational data;
• human authority for safety-critical decisions; and
• post-deployment monitoring for drift and anomalous recommendations.

An AI system may recommend a layout, but the responsible engineer remains accountable for sign-off, evidence, and compliance. Automation should increase analytical capacity without eliminating professional judgement.

Emerging Device and System Technologies

Flexible and Printed Electronics

Flexible electronics use thin substrates, organic semiconductors, nanomaterials, printed conductors, or thin-film devices to create bendable sensors, displays, wearable systems, and conformal antennas. Their advantages include low mass, mechanical conformity, and potentially large-area manufacturing. Their challenges include lower or variable mobility, moisture and oxygen sensitivity, barrier integrity, contact resistance, mechanical fatigue, and limited repairability. Reliability testing must include bending radius, cyclic flexing, stretching, washing where relevant, abrasion, humidity, and body-temperature exposure.

Neuromorphic Devices

Neuromorphic systems imitate selected principles of biological information processing through event-driven computation, distributed memory, and neuron-like dynamics. Resistive-switching devices, phase-change devices, spintronic elements, and specialized CMOS circuits may act as synaptic or neuronal components. Important figures of merit include energy per event, endurance, retention, variability, noise tolerance, learning stability, and system-level accuracy. Device variability can be a resource for stochastic computing but is a defect when uncontrolled. Hardware–algorithm co-design is essential because the algorithm must accommodate analogue non-idealities and limited precision.

Chiplets and Advanced Integration

Chiplets divide a system into smaller functional dies connected through an advanced package. This can improve reuse, yield, heterogeneous integration, and time to market. It can also introduce die-to-die latency, interconnect power, thermal hotspots, package warpage, interface-security concerns, and new test requirements. The chiplet ecosystem requires standardized interfaces, known-good-die strategies, package-level verification, and clear responsibility for defects at die and integration boundaries.

Quantum Devices

Quantum devices exploit quantum states for sensing, communication, or computation. Their engineering difficulties include extreme sensitivity to noise, temperature, vibration, electromagnetic interference, fabrication variation, and control errors. Reliability cannot be described only by classical component lifetime; coherence, gate fidelity, calibration stability, error rates, and cryogenic infrastructure are also relevant. Quantum technology should therefore be evaluated by complete system utility rather than by a single laboratory metric.

Sustainability, Ethics, and Governance

Sustainable electronics considers material extraction, fabrication energy, water use, hazardous substances, transportation, operational energy, repair, reuse, recycling, and disposal. Semiconductor fabrication and cooling can be resource-intensive, while frequent product replacement increases embodied impact. Life-cycle assessment helps compare alternatives, but results depend on system boundaries and data assumptions. Design strategies include lower-power operation, longer service life, modular replacement, software support, recycled or less hazardous materials, and recovery of valuable constituents.

Ethical engineering includes safety, privacy, accessibility, truthful performance claims, worker welfare, and avoidance of planned premature obsolescence. Connected devices can collect sensitive operational or personal data. Predictive maintenance can affect employment, warranty decisions, or access to service if its predictions are opaque or biased. Engineers should document limitations, protect data, provide human review, and avoid presenting probabilistic outputs as certainty.

Governance assigns authority and accountability. A practical governance record identifies requirements, hazards, assumptions, datasets, test evidence, configuration baselines, approval gates, deviations, and residual risks. Safety-critical systems require independent review and change control. Cybersecurity must be included in reliability analysis because malicious modification, counterfeit components, compromised firmware, or denial of service can produce functional failure even when hardware is physically intact.

Illustrative Case Study: Reliability Selection for an Industrial Motor Drive

Background

An illustrative manufacturer is developing a variable-speed drive for a 15-year industrial service life. The drive operates continuously, experiences daily load changes, and is installed in a warm, dusty enclosure. Two power-module options meet the electrical rating.

Situation and Problem

Module A has a lower purchase price and a shorter, familiar qualification history. Module B costs more but has lower thermal resistance, a longer stated supply horizon, and an established second source. The procurement team prefers A, while the reliability team is concerned about junction-temperature cycling and future availability.

Analysis

The team first establishes the mission profile: load distribution, switching frequency, ambient temperature, enclosure airflow, maintenance intervals, and acceptable downtime. Thermal simulation shows that A operates closer to its maximum junction temperature during peak load. FMEA identifies solder fatigue and thermal overstress as high-consequence failure modes. Supply analysis identifies A as single-source with a long replacement lead time.

A lifecycle model includes acquisition, thermal redesign, qualification, expected replacement, inventory, downtime, and end-of-life risk. The team does not use the supplier’s quoted lifetime without checking test conditions. It conducts representative power-cycling and temperature-cycling tests, examines failed samples, and evaluates whether B’s second source is genuinely interchangeable at the electrical, mechanical, firmware, and qualification levels.

Strategy and Solution

The selected design uses Module B, improved heatsinking, temperature sensing, conservative current derating, and a serviceable power subassembly. A second-source approval plan and incoming inspection are established. A condition-monitoring algorithm tracks thermal cycles and abnormal conduction loss, but maintenance decisions remain subject to engineering thresholds and human review.

Outcome

The decision increases initial cost but reduces predicted thermal stress, improves repair planning, and lowers supply interruption exposure. The organization also gains a documented evidence chain connecting mission profile, test results, design margins, supplier controls, and maintenance actions.

Learning Points

The case demonstrates that reliability is an architectural and economic decision, not merely a component rating. It also shows why AI or monitoring cannot compensate for poor thermal design, and why supply continuity and maintainability belong in the original specification.

Challenges and Common Mistakes

Frequent mistakes include treating maximum ratings as operating targets, extrapolating accelerated tests outside their physical validity, ignoring common-cause failures, using MTBF as a guarantee for an individual unit, selecting AI models by accuracy alone, and confusing regulatory compliance with safety. Other errors include underestimating connector and packaging failures, failing to preserve traceability after design changes, and evaluating sustainability only through operating power while ignoring embodied materials and disposal.

Reliability data are often incomplete because field failures are rare, reporting is inconsistent, and products operate under different conditions. Engineers should state confidence intervals, censoring assumptions, sample limitations, and uncertainty. When evidence is weak, the appropriate response is not false precision but additional testing, conservative design, or explicit residual-risk acceptance.

Best-Practice Decision Workflow

A defensible workflow is to define mission and failure consequences; identify hazards and critical functions; construct a stress–mechanism–failure model; compare alternatives using lifecycle criteria; design for margin, testability, manufacturing, and repair; qualify under representative conditions; monitor field evidence; and update the model through controlled change management. At each gate, technical, manufacturing, supply-chain, sustainability, cybersecurity, and ethical stakeholders should be able to challenge assumptions.

Current Directions

Important directions include physics-informed machine learning, automated reliability-aware layout, in-situ health monitoring, advanced thermal materials, heterogeneous integration, recyclable substrates, low-power edge intelligence, and standardized chiplet interfaces. Progress will depend less on isolated device demonstrations than on reproducible manufacturing, reliable packaging, test access, secure software, lifecycle evidence, and responsible governance. The central engineering question is not whether a technology is novel, but whether it delivers dependable value under realistic constraints.

Chapter Summary

Reliability is the probability of successful function under specified conditions and time, while availability additionally depends on repair and restoration. Electronic failures arise from semiconductor degradation, packaging fatigue, thermal stress, contamination, mechanical loading, manufacturing defects, and system interactions. Reliability engineering uses mission profiles, FMEA, fault trees, derating, qualification, accelerated testing, and root-cause analysis.

Engineering decisions should include lifecycle cost, manufacturability, maintainability, supply-chain resilience, sustainability, cybersecurity, and regulatory evidence rather than purchase price alone. AI-assisted EDA, digital twins, and predictive analytics can improve design and maintenance, but require traceability, physical validation, uncertainty management, and human accountability. Flexible, neuromorphic, chiplet, and quantum technologies introduce new performance opportunities and new reliability models. Ethical and professional practice requires transparent claims, privacy protection, safety governance, environmental responsibility, and controlled management of residual risk.

Key Terms

Reliability: Probability that an item performs its required function for a specified time and condition.

Availability: Probability that a system is operational when required.

Maintainability: Ability to be restored to service efficiently after failure.

MTBF: Mean time between failures for a repairable item.

MTTR: Mean time to repair or restore service.

Bathtub curve: Model showing early-life, useful-life, and wear-out failure regions.

Derating: Operating a component below its maximum specified stress.

FMEA: Bottom-up analysis of failure modes, causes, effects, and controls.

Fault-tree analysis: Top-down logical analysis of combinations leading to a hazardous event.

Qualification test: Controlled test providing evidence that a design and process meet requirements.

Accelerated testing: Testing under increased stress to obtain reliability evidence more quickly.

Electromigration: Current-induced movement of metal atoms in interconnects.

Thermal cycling: Repeated temperature variation that can cause mechanical fatigue.

Lifecycle cost: Total cost from acquisition through operation, maintenance, failure, and disposal.

Digital twin: Data-connected computational representation of a physical asset or process.

Predictive maintenance: Maintenance scheduled using estimated future condition or failure risk.

Chiplet: Functional semiconductor die integrated with other dies in an advanced package.

Neuromorphic electronics: Hardware designed to implement brain-inspired, event-driven computation.

Common-cause failure: Multiple failures produced by one shared cause.

Residual risk: Risk remaining after controls and mitigation are applied.

Review Questions

Short Answer Questions

1. Distinguish reliability, availability, maintainability, and quality.

2. Why can accelerated testing produce misleading results when its stress changes the failure mechanism?

3. What is the purpose of derating in electronic design?

4. State two limitations of AI-based predictive maintenance.

5. Why are chiplet systems subject to package-level reliability concerns?

Long Answer Questions

1. Explain the bathtub curve and relate each region to appropriate engineering actions.

2. Describe how FMEA and fault-tree analysis complement each other in a reliability programme.

3. Discuss lifecycle cost and supply-chain resilience as criteria in component selection.

4. Explain the reliability challenges of flexible and neuromorphic electronics.

5. Develop a governance framework for responsible use of AI-assisted EDA in a safety-critical product.

Analytical / Application Questions

1. A converter has a high calculated MTBF but requires two days for field repair. Analyze its availability and recommend architectural improvements.

2. Compare two power modules when one has lower price and the other has lower thermal resistance and a second source. Define a multi-criteria decision method and identify the evidence required.

3. Design a qualification and monitoring strategy for a wearable flexible sensor exposed to bending, humidity, sweat, and intermittent wireless operation.

22. Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes reliability from availability?

A. Reliability concerns repair time, whereas availability concerns only initial quality.

B. Reliability is probability of functioning for a specified period; availability also includes restoration after failure.

C. Reliability applies only to software, whereas availability applies only to hardware.

D. Reliability and availability are identical measures under all operating conditions.

Correct Answer: B. Reliability is probability of functioning for a specified period; availability also includes restoration after failure.

Question 2

A qualification test is most likely to give misleading lifetime evidence when it:

A. Uses representative mission profiles and failure analysis.

B. Includes independent samples and documented conditions.

C. Uses a stress level that activates a different physical failure mechanism.

D. Combines testing with field data and confidence analysis.

Correct Answer: C. Uses a stress level that activates a different physical failure mechanism.

Question 3

Which technique begins with an undesirable system event and decomposes it through logical AND and OR relationships?

A. FMEA

B. Fault-tree analysis

C. Design for assembly

D. Life-cycle assessment

Correct Answer: B. Fault-tree analysis

Question 4

A design team selects a component solely because it has the lowest purchase price. Which important consideration has most clearly been omitted?

A. Lifecycle cost and failure consequence

B. Nominal logic polarity

C. The existence of semiconductor physics

D. The definition of voltage

Correct Answer: A. Lifecycle cost and failure consequence

Question 5

What is a major risk of using an AI optimizer for circuit layout without physical constraints?

A. It necessarily eliminates all process variation.

B. It may optimize a stated objective while violating reliability or manufacturability requirements.

C. It cannot process any design data.

D. It guarantees explainable decisions.

Correct Answer: B. It may optimize a stated objective while violating reliability or manufacturability requirements.

Question 6

Which feature is essential for a useful digital twin?

A. Aesthetic three-dimensional visualization only

B. A connection between a computational representation and relevant physical or operational data

C. Removal of all uncertainty from measurements

D. Replacement of every physical test

Correct Answer: B. A connection between a computational representation and relevant physical or operational data

Question 7

In a chiplet-based system, which issue is especially associated with integration boundaries?

A. Die-to-die interface integrity and package thermal interaction

B. Elimination of all testing requirements

C. Guaranteed independence from packaging

D. Inability to reuse functional dies

Correct Answer: A. Die-to-die interface integrity and package thermal interaction

Question 8

Why should neuromorphic-device evaluation include variability and endurance?

A. Because device non-idealities can affect learning stability and long-term function.

B. Because neuromorphic systems never use electronic devices.

C. Because endurance is relevant only to mechanical components.

D. Because variability always improves accuracy.

Correct Answer: A. Because device non-idealities can affect learning stability and long-term function.

Question 9

Which action best reduces single-source supply-chain risk?

A. Removing incoming inspection

B. Using a qualified second source and designing for substitution where feasible

C. Increasing the nominal operating stress

D. Ignoring supplier continuity information

Correct Answer: B. Using a qualified second source and designing for substitution where feasible

Question 10

A predictive-maintenance model performs well during development but poorly after deployment because operating conditions have changed. This is an example of:

A. Distribution shift

B. Derating

C. Electromigration

D. Package delamination

Correct Answer: A. Distribution shift