The Day the Yield Report Came Out, My Boss Simply Asked: "How Long Will This Thing Last?"
A few years ago, we had a new product that had just entered mass production. The yield numbers were excellent, with a Cpk hitting 1.08 and DPMO at only 6210. Everyone was thrilled. However, during a meeting, the boss looked at the reliability prediction report with a somewhat stiff expression. Pointing to a number on the report, he slowly said: "Good yield is fundamental, but how long is this product of yours expected to last? If customers buy it and it breaks after a year, how many complaints should we prepare for?" The room instantly fell silent. Everyone exchanged glances because that report was calculated using MIL-HDBK-217, and the numbers looked... well, very unoptimistic.
What's the Problem? What Exactly Does MIL-HDBK-217 Predict?
To be frank, reliability prediction, at its core, is about "estimating if a product will fail, and roughly when." In our factory's early days, we always used the MIL-HDBK-217 standard. This standard was developed by the U.S. military, initially for military electronic equipment. You know military products—they're required to be rugged, durable, and functional under various extreme conditions. Therefore, its prediction method considers environmental factors like temperature, voltage, vibration, and humidity very, very stringently.
The key point is, when you apply the MIL-HDBK-217 formula to a consumer electronic product, such as a mobile phone or home appliance, the resulting failure rate will typically be so high it makes you question everything. This is because it assumes your product is used on a battlefield, whereas our typical products are used in offices or homes. It's like demanding a sedan meet the durability standards of a tank; naturally, it will fail every calculation.
How Is It Done in Practice? Telcordia to the Rescue!
Later, we discovered that if you truly want to predict the reliability of consumer products, using Telcordia SR-332 (formerly known as Bellcore) is more reasonable. Telcordia is a standard developed by the U.S. telecommunications industry, primarily for communication equipment. While communication equipment also demands stability, its operating environment is typically a server room, which is relatively less extreme.
In other words, Telcordia's prediction model is closer to commercial environments. It also considers ambient temperature, component type, operating voltage, and even the quality grade of components (Commercial Grade, Industrial Grade). For example, if a capacitor's predicted failure rate under MIL-HDBK-217 is 10 FIT (Failures In Time, failures per billion hours), it might only be 3 FIT under Telcordia. This number is much more realistic and more acceptable to both management and customers.
The Most Common Pitfall: Don't Blindly Copy!
I remember once, we had a meeting with a new supplier, and they also presented a MIL-HDBK-217 report. I asked them, "For this board, are you using commercial-grade components or military-grade? What is the ambient temperature setting?" The supplier's engineer stammered, saying they just ran the software directly, using all default parameters.
To put it plainly, choosing a prediction method is one thing, but the parameter settings within it are the key! Many simply and naively apply the software's default values, resulting in figures that are either overly pessimistic or overly optimistic, completely failing to reflect the real situation.
- Choose the correct component grade: Are you using general Commercial Grade components, Industrial Grade, or even Military Grade? These will significantly impact the failure rate.
- Ensure accurate ambient temperature: What is your product's actual operating temperature? Server room? Office? Or outdoors? A 10-degree difference in the input temperature parameter can lead to a several-fold difference in the failure rate.
- Don't randomly input component derating: The derating of capacitors and resistors must also be considered. Are they being overstressed?
One Thing You Can Do Today
Review your reliability prediction report and confirm that the parameter inputs are reasonable.