HALT and HASS Planning

HALT and HASS Planning

HALT and HASS Planning

What the two parts do

A highly accelerated life test (HALT) steps a few prototypes through cold, heat, random vibration and the combined environment until they stop working and then until they break, to find the weak links of the design and the margins beyond the specification. A highly accelerated stress screen (HASS) or an environmental stress screen (ESS) then runs every production unit through a shorter, milder version of those stresses to precipitate and remove the latent workmanship and part defects before shipment. The first part of this tool organizes the HALT results (limits, margins, the HASS levels they imply, the step-stress profiles, the sample-size logic); the second quantifies a screen with the mathematics of MIL-HDBK-344A: how many latent defects it removes, how many escape, what that means for the field MTBF, what the screen costs and saves, how much fatigue life it consumes, and how to prove its strength with seeded defects.

HALT limits, margins and HASS levels

Four limits come out of each HALT stress: the lower and upper operating limits (LOL, UOL), where the product stops functioning but recovers when the stress is relaxed, and the lower and upper destruct limits (LDL, UDL), where it fails permanently; vibration gives an operating limit (VOL) and a destruct limit (VDL) in Grms. The operating margin is the distance from the specification limit to the operating limit, the destruct margin the distance from the operating to the destruct limit. Hobbs (2000) recommends increasing the margins by design until the limits are set by the fundamental limits of the technology, and then placing the HASS precipitation screen beyond the operating limits, part way toward the destruct limits, where the product need not function but the latent defects are driven to failure, and the detection screen inside the operating limits, where the product must function so that the precipitated defects can be seen by the monitoring. The tool places the precipitation level at the operating limit plus a planner's fraction of the destruct margin, and the detection level at the specification plus a fraction of the operating margin; both fractions default to one half and are entered, because no published rule gives them as numbers. The levels are then confirmed by a proof of screen (seeded defects, below) and a safety of screen (the fatigue life the screen consumes, below).

The step-stress tables start at the entered temperatures and vibration level and step by the entered increments with one dwell per step until the destruct limit is reached, marking the steps beyond the operating limit. The combined profile cycles between the precipitation temperatures at the chamber's product ramp rate, with the vibration level raised by one step each cycle. These are planning drawings: in the chamber the steps are usually halved as a limit is approached and the dwell is set by the time the product's internal temperatures take to stabilize.

Sample size. HALT is a small-sample test aimed at design weaknesses that every unit shares. With n units the probability of seeing at least one unit carrying a weakness present in a fraction p of the population is 1 − (1 − p)ⁿ, the smallest prevalence revealed with confidence C is 1 − (1 − C)^1/n, and the units needed for a prevalence p at confidence C are ln(1 − C)/ln(1 − p). Four units see a 20 % weakness with 59 % probability; lot and workmanship defects are the business of HASS on every unit.

Screening strength per MIL-HDBK-344A

The handbook models a screen as two stages (Appendix A). The precipitation efficiency PE is the probability that the stress turns a latent defect into a detectable (patent) one; it grows with the stress duration as PE = 1 − exp(−k t), where k is a stress constant for the screen type (Procedure C step 1, from RADC-TR-86-149):

• temperature cycling: k = 0.0017 (ΔT + 0.6)^0.6 [ln(RATE + e)]³, t = number of cycles, ΔT = T_max − T_min in °C, RATE the average rate of change of the hardware temperature in °C/min (Table 5.15);
• constant temperature: k = 0.0017 (ΔT + 0.6)^0.6, ΔT = |T − 25 °C|, t in hours after thermal equilibrium (Table 5.17);
• random vibration: k = 0.0046 G^1.71; swept sine: k = 0.000727 G^0.863; fixed sine: k = 0.00047 G^0.49; G in Grms at the hardware, t in minutes (Tables 5.14 and 5.16). A random-vibration PE is multiplied by the axis sensitivity factor, the fraction of the vibration-sensitive defects that respond to the axis excited.

The detection efficiency DE is the probability that a precipitated defect is detected, isolated and removed; the handbook takes it as the product of three factors (Procedure C step 2): the type of test (functional only 0.5 to 0.8, functional and parametric 0.8 to 1.0), the environment during test (ambient only 0.2 to 0.6, concurrently with the stress 1.0) and the ability to isolate and remove the defect without introducing another (0.8 to 1.0). The screening strength is SS = PE × DE. The handbook's own example, 4 cycles at 5 °C/min over 100 °C (PE 0.6027) and 5 minutes at 5 Grms (PE 0.303) with DE 0.9 × 1.0 × 0.95 = 0.855, gives SS 0.5153 and 0.259; it is the default of this page. The tables of the handbook are reproduced by the equations to the printed digit, which the tool's test script verifies.

The constants were fitted to screens of the handbook's era (assemblies and units, up to about 180 °C of range, 30 °C/min and 10 Grms). A HASS placed from HALT limits may lie beyond that range; the equations then extrapolate, and the proof of screen is the check.

Defect flow, yield and escapes (Procedure A)

Latent defects are introduced at a density D_IN (defects per unit). The handbook proportions them into a population that responds to thermal screens and one that responds to vibration (historically 80 % and 20 %, 4.4.4.1), because the two screen types reach different defect types (Table 4.3). Each screen removes the fraction SS of the population it reaches, so along a flow D_OUT = Σ D_i Π (1 − SS_j) (Figure 5.1); the tool runs the thermal screens on the thermal population and the vibration screen on the other. The yield exp(−D_REMOVED) is the probability that a unit passes the screen with no fallout (eq. A-7), and 1 − exp(−D) is the probability that a unit carries at least one latent defect (eq. A-5), before and after screening. The overall screening strength of the program is D_REMOVED/D_IN.

Allowable remaining defects. Procedure A1 step 5 ties the screen to the field requirement. The failure rate the latent defects may contribute is what the required MTBF allows beyond the inherent (limiting) MTBF, divided by a design safety margin (1.5 to 2): FR ≤ (1/MTBF_req − 1/MTBF_inh)/SM. In a field period t the remaining defects precipitate with the field screening strength 1 − exp(−k t), where k is the field precipitation rate (1/500 to 1/2000 per hour when the field stresses are not known), so the defects that may remain at shipment are D_REMAINING = FR t / (SAF [1 − exp(−k t)]), with SAF the stress adjustment factor between factory and field stress levels (Procedure B2). The required screening strength is then (D_IN − D_REMAINING)/D_IN, and the tool says whether the planned screens meet it.

Defect density from a part list. Procedure B estimates D_IN = Σ n_i p_i (eq. A-1) from the part quantities by class and quality grade and the baseline defect densities of Table 5.1 (PPM per part at 6 Grms and 6 °C/min; assembly characteristics 25 PPM per part, lead, terminal, wire or board, solder connections 5 PPM each). The tool's part-list box takes one line per class and grade. The handbook's Tables 5.2 to 5.13 give field-environment densities for the stress adjustment factor; they are not transcribed here, and the sample system matrix of its Figure 5.7 is not reproducible from the printed tables, so Table 5.1 alone is used.

Cost. Procedure A3 counts the test, screen, rework and field-failure costs. Per unit, without screening every latent defect that precipitates in the field period is a field failure; with screening the screen, the rework of the factory fallout and the escapes that still precipitate are paid. The difference is the saving the screen earns at the entered unit costs.

Safety of screen (Procedure A4) and MIL-HDBK-2164A

A screen must not use up the product's fatigue life. The handbook compares damage indices D = N S^B of the screen and of the useful life, with N the cycles or minutes, S the temperature range or Grms and B a fatigue exponent, 2.5 for thermal cycling and 6.4 for random vibration of solder joints; the ratio is the fraction of life one screen consumes. Its example, 7,300 life cycles of 30 °C against 50 screen cycles of 120 °C, gives 21.9 %; 2 × 10⁶ minutes at 1 Grms against 5 minutes at 6 Grms gives 23.8 %. The index is a Coffin-Manson / Basquin form; it is misleading for short-life items and must be summed over every repeat of the screen after rework. MIL-HDBK-2164A applies the same idea to its 6 Grms (0.04 g²/Hz) screen: all acceptance and diagnostic vibration on one unit should not exceed 20 equivalent minutes at full level, with the equivalent time scaling as (PSD/0.04)³, i.e. (G/6)⁶; the tool prints the planned screen's equivalent minutes.

Proof of screen with seeded defects

The handbook defines the measured screening strength as the fallout divided by the defects initially present (4.4.4.2.3). Seeding known defects (loose hardware, lifted leads, cracked joints) into proof units and counting how many the screen finds gives k of n; the point estimate is k/n and the binomial (Clopper-Pearson) bounds show what the test demonstrates. Finding all n seeded defects demonstrates SS > (1 − C)^1/n at confidence C, so to show SS above a target SS₀ the number seeded must satisfy SS₀ⁿ ≤ 1 − C: 22 for 0.90 at 90 %.

Published screens offered as presets

MIL-HDBK-344A Table 4.5 (R&M 2000 initial regimen): assemblies −54 to +85 °C at 30 °C/min minimum, 25 cycles; units −54 to +71 °C at 5 °C/min, 10 cycles, random vibration 6 Grms 100 to 1000 Hz, 10 minutes per axis. MIL-HDBK-2164A: thermal cycling −54 to +71 °C at 10 °C/min minimum, 10 cycles minimum, with random vibration at 0.04 g²/Hz (6 Grms, 20 to 2000 Hz) for 5 minutes before and 5 defect-free minutes after the thermal cycling, and a 40-hour defect-free thermal-cycling period within 80 hours. The presets fill the screen fields with these values.

Limits of the tool

The precipitation constants are the handbook's industry averages and are to be refined from factory fallout (Procedure D); the detection factors are judgments within the handbook's ranges; the 80/20 split is historical. The flow treats two defect populations and the screens in the order thermal, constant temperature, vibration on one assembly level, not a multi-level integration tree (the handbook's Figure 5.2 is one). The HASS levels are placed by the planner's fractions, not by a published rule, and the damage-index exponents are the handbook's solder values unless entered.