In the field of semiconductor reliability testing, particularly in HTOL (High Temperature Operating Life), the sample size of 77 is the "golden number" derived from statistical principles, industry standard practices, and a precise cost-benefit trade-off. The core reason is to demonstrate, under the premise of "zero failures," that the chips achieve an extremely high reliability target (typically a very low FIT rate) with high statistical confidence at a reasonable cost.
       1. The Roots of Statistics: Chi-Square Distribution and Confidence Levels
       The goal of HTOL is to verify that the failure rate of the chip is extremely low within its expected lifespan. Due to limited testing time and sample size, the actual MTTF (Mean Time to Failure) cannot be measured directly, and only the confidence lower bound of MTTF can be estimated through statistical inference. The most commonly used method is the Chi-Square Distribution formula:MTTF ≥ (2 × N × t × AF) / χ²(α, 2)

N: Number of test samples (a calculated figure)
t: Test duration (typically 168 hours or 1000 hours)
AF: Acceleration factor (calculated by the Arrhenius model based on test temperature and actual operating temperature)
χ²(α, 2): The chi-square value, where α is the risk rate and 2 is the degrees of freedom (number of failures + 2)
Confidence Level = 1 - α
       2. Derivation Process
       Set the goal: Prove that the chip's FIT (Failures In Time per billion device hours) < 1, i.e., MTTF > 10⁹ hours.
       Set conditions:
· Number of failures = 0 (Zero failures is a fundamental requirement)
· Confidence level = 60% (corresponding risk rate α = 0.4)
· Consult the chi-square distribution table, χ²(α=0.4, df=2) ≈ 1.83
· Test duration t = 1000 hours (common HTOL test duration)
· Acceleration Factor (AF): This is a variable that depends on temperature. In a typical scenario, assuming an HTOL test temperature of 125°C and an actual operating temperature of 55°C with an activation energy (Ea) of 0.7 eV, the calculated AF ≈ 77.8.
Calculate the quantity N: 10⁹ ≤ (2 × N × 1000 × 77.8) / 1.83
Solving: N ≥ (10⁹ × 1.83) / (2 × 1000 × 77.8) ≈ 77.05
It requires 77 samples to pass a zero-failure test at 125°C for 1,000 hours, ensuring a 60% confidence level that the chip's failure rate in a 55°C operational environment is below 1 FIT (i.e., MTTF exceeding 1 billion hours).

3. Industry standard practice: 60% confidence balance
60% confidence level: widely recognized by the industry as the best balance between statistical rigor and testing costs. It provides sufficient credibility while controlling costs.
Higher confidence level (such as 90%): More samples (such as 184) are required, but it will significantly increase testing costs (expensive chips, test benches, machine time, etc.).
Interchangeability of samples and time: Less samples can be tested for a longer time (such as testing 38 samples for 2000 hours), or more samples can be tested for a shorter time (such as testing 77 samples for 1000 hours) to achieve the same statistical power.
4. Cost effectiveness: The most cost-effective solution
Statistical inevitability: 77 samples are derived from the chi square distribution formula, and under standard assumptions such as "zero failure", "60% confidence level", "1000 hour test", and "typical acceleration factor", they are the minimum integer sample size that satisfies the 1 FIT failure rate target.
Industry convention: 60% confidence is the tacit understanding and optimal balance point reached by the industry between statistical rigor and testing costs.

This article is reprinted from the internet. If there is any infringement, please contact us to delete it. Thank you!