Datasheets, Component Selection, Tolerances & Reliability

Choosing parts that work in production, not just on the bench. Learn how to read datasheets (absolute maximum vs recommended vs electrical characteristics, typ/min/max, test conditions, graphs), selecting op-amps, resistors, capacitors, inductors, diodes and transistors, tolerance stacking (worst case, RSS, Monte Carlo), temperature and voltage derating, ceramic capacitor DC-bias loss, electrolytic life (Arrhenius), resistor power and pulse ratings, reliability metrics (FIT, MTBF, bathtub curve), second sourcing, lifecycle and counterfeit risk — with Python computing derating, capacitor life, DC-bias loss and system FIT/MTBF.

A design that works with the parts on your bench can fail with the next reel, the next temperature or the next year. Robust designs are built from datasheet limits, not typical values, with margins for tolerance, temperature, ageing and supply variation.

Component selection also includes availability, cost, lifecycle and reliability — engineering decisions as important as circuit topology.

An analogy

Reading a datasheet is like reading a car's specifications: the 0–100 km/h time (typical) is what it does on a good day with a professional driver; what matters for safety is the guaranteed braking distance (min/max) under the stated conditions.

Passives

Selection notes:

Actives

Op-amps: supply and input/output ranges, V_OS/drift, I_B, GBW, slew rate, noise, stability with load, package. Transistors/MOSFETs: voltage and current ratings with margin, SOA, R_DS(on) at your V_GS and temperature, gate charge, thermal resistance. Diodes: reverse voltage, recovery time, leakage at temperature. References and ADCs: accuracy over temperature and life. Prefer parts with multiple sources and long lifecycle for products.

Tolerance analysis

Worst-case analysis guarantees function but can over-design; RSS and Monte Carlo predict yield. Consider initial tolerance, temperature drift, ageing (e.g. resistor drift after 1000 h at 70 °C) and supply variation. Sensitivity analysis shows which components need tight tolerance — spend precision only where it matters.

Reliability engineering

Failure rates follow the bathtub curve: infant mortality (screened by burn-in), constant random failures (FIT-based MTBF), wear-out (electrolytics, fans, connectors, solder fatigue). Prediction standards (e.g. SN 29500, MIL-HDBK-217 legacy) estimate FIT. Improve reliability with derating, lower temperatures, fewer parts and connectors, robust protection, and qualification (HALT/HASS, temperature cycling, humidity). Manage supply-chain risk: authorised distributors, second sources, lifecycle monitoring.

Formulas

Worked example: MTBF

Given: Total 200 FIT

Find: MTBF.

Answer: 5 × 10⁶ h ≈ 571 years

Worked example: Capacitor life

Given: 3000 h at 105 °C, used at 75 °C

Find: Life.

Answer: 24 000 h

Worked example: Resistor power margin

Given: 0.1 W dissipation, 0603 rated 0.1 W

Find: Verdict.

Answer: Use 0805/1206 (≥ 2× margin)

Datasheet reading checklist

  Absolute maximum ratings → never exceed, not operating conditions
  Recommended operating conditions → design within these
  Electrical characteristics → use MIN/MAX (over temperature!), note test conditions
  Typical graphs → trends, not guarantees (read axes and conditions)
  Application info / layout → often essential for stability
  Package, thermal, ordering, lifecycle (NRND/EOL), errata

Typical values are for understanding; min/max values over temperature are for design. Always check which conditions (supply, load, temperature) each spec applies to — a spec at 25 °C tells you little about 85 °C.

Derating, capacitor life, DC bias and FIT

# Component selection & reliability: derating, capacitor life, ceramic DC bias, FIT/MTBF
V_rail, rating = 12.0, 25.0
print(f"capacitor voltage derating: {V_rail/rating*100:.0f}% of rating (target <= 50-80%)")
L0, T0 = 2000.0, 105.0                          # hours at rated temperature
for T in (85.0, 65.0):
    print(f"electrolytic at {T:.0f} C: life ~ {L0*2**((T0-T)/10):.0f} h ({L0*2**((T0-T)/10)/8760:.1f} years)")
C_nom, loss = 10e-6, 0.55                         # X5R 10 uF 0603 at 5 V bias loses ~55%
print(f"10 uF X5R 0603 at 5 V DC bias: ~{C_nom*(1-loss)*1e6:.1f} uF effective")
parts = {"op-amp": 5, "LDO": 10, "MOSFET": 8, "resistors x40": 40 * 0.2, "ceramics x30": 30 * 0.3, "electrolytic": 20, "connector": 15}
fit = sum(parts.values())
print(f"total FIT {fit:.0f} -> MTBF {1e9/fit/8760:.0f} years; P(fail in 5 years) = {(1-2.718281828**(-fit*1e-9*8760*5))*100:.2f}%")
# capacitor voltage derating: 48% of rating (target <= 50-80%)
# electrolytic at 85 C: life ~ 8000 h (0.9 years)
# electrolytic at 65 C: life ~ 32000 h (3.7 years)
# 10 uF X5R 0603 at 5 V DC bias: ~4.5 uF effective
# total FIT 75 -> MTBF 1522 years; P(fail in 5 years) = 0.33%

A 25 V capacitor on a 12 V rail is at 48% of rating — good practice. An electrolytic rated 2000 h at 105 °C lasts under a year at 85 °C but ~3.7 years at 65 °C, so cooler placement matters more than you think. A small 10 µF X5R part may give only ~4.5 µF at 5 V. Summing FIT rates gives system MTBF and the chance of a failure over the product life.

Common mistakes

Applications

Key takeaways

Going further