Packaging and Board Reliability

The package around a die, the interconnect between that package and its board, the board itself, and the coatings, platings and bonds that hold the assembly together. A part number carries no rating of its own in this class. The result is a cycle count between two temperatures, or a board strain against a stated failure criterion, and the same leadless chip carrier failed at 50 cycles over a 235 C delta and at 170 cycles over a 145 C delta [1].

Three mechanisms produce almost every entry below. The first is low-cycle fatigue driven by mismatched thermal expansion, which sets the cycle counts in the first table and is the reason an underfill whose expansion coefficient is wrong makes a joint worse rather than better: a 256 I/O ceramic ball grid array underfilled at 75 ppm per K first failed at 931 cycles from 0 to 100 C, against 1980 cycles for the same package with no underfill at all [6]. The second is board strain, which fractures the intermetallic layer of a solder joint or craters the pad beneath it at strains an electrical continuity test does not see; on a 1521 I/O flip-chip BGA acoustic emission put brittle fracture at 860 microstrain and pad cratering at 1250, where daisy-chain continuity on the same specimen read 2800 to 3500 [5]. The third is that the package participates in effects it was never qualified against, from tin whisker growth under a pure tin plating [8] to stopping the ions of a heavy-ion beam before they reach the die [13].

Packages and interconnects with a thermal cycling result

Every row is a chamber result in dry gas. None is a combined-environment exposure: no CO2 at Mars ambient pressure, no dust, no radiation, and dwells of 10 to 45 minutes rather than a Martian diurnal hold, so creep-dominated damage is under-represented throughout [1].

Package or interconnect	Assembly	Used by	Status	Source
LCC20 leadless chip carrier	Engineering camera	spirit, opportunity	Failed, redesigned	[1][2]
L-shaped-lead package	Engineering camera	spirit, opportunity	Ground test	[1]
Looped haywire interconnect	Engineering camera	spirit, opportunity	Ground test	[1]
Engineering camera package, redesigned	Engineering camera	curiosity	Qualified	[1][2]
Type X platinum resistance thermometer	Bonded temperature sensor	spirit, opportunity	Flown, failed	[2][3]
HRTS-5760-B-U-0-12, Honeywell	Bonded temperature sensor	curiosity	Flown 2012	[3]
FB500 flip-chip test board	Daisy-chain coupon	none	Ground test	[2]
NVMCAM flash memory slice	Rover compute element	curiosity, perseverance	Qualified	[4]
PIXL Low Voltage Control Module	Instrument electronics	perseverance	Qualified below its own cold limit	[7]
Motor encoder chip-on-board	Actuator feedback	curiosity	Flown 2012	[9]
HTCC alumina, co-fired platinum	Venus surface electronics	none	Ground test	[11]

Ratings and qualification results

• LCC20 leadless chip carrier, manufacturer not named. Ratings: no part number, solder alloy or board stackup is published for it [1]. Qualification: 50 thermal cycles to a cracked interconnect from -120 to +115 C, a 235 C delta, at 5 C per minute with 45 minute hot and 10 minute cold dwells, against a Mars Exploration Rover requirement of 200 cycles [1]. The same package type reached 170 cycles on the MSL summer profile of -105 to +40 C, a 145 C delta at 15 minute dwells, and passed 2010 cycles only after its mounting was redesigned [2]. Sample size is not stated for either figure, so neither is a lifetime distribution.
• L-shaped-lead package, manufacturer not named. Ratings: an alternate lead geometry evaluated against the leadless carrier in the same camera assembly [1]. Qualification: 121 cycles to failure at a 205 C delta. The source converts that to 241 MSL-equivalent cycles using an acceleration factor it never states, so the equivalent count cannot be checked [1].
• Looped haywire interconnect, manufacturer not named. Ratings: a compliant wire interconnect evaluated in the same camera assembly [1]. Qualification: 200 cycles survived at a 205 C delta, converted by the same unstated factor to 400 MSL-equivalent cycles [1].
• Engineering camera package, redesigned, JPL. Ratings: qualified against a JPL design principle of three times the cycles expected in the mission, not against a part specification [2]. Qualification: 200 cycles from -120 to +115 C for the 90-sol Mars Exploration Rover prime mission, and 2010 cycles for Curiosity, run as 470 summer cycles from -105 to +40 C plus 200 winter cycles from -130 to +15 C per year for three years, with the camera still functional at the end [1]. Both profiles come from Ames global circulation model predictions for a landing site not yet chosen rather than from measured Mars surface temperatures, and the winter cold end is below the -65 to +125 C range standard military-specification packages are built for [1]. The qualification covers the package and its interconnect; nothing in it says the detector still meets its noise or responsivity specification afterward.
• Type X platinum resistance thermometer, B.F. Goodrich. Ratings: a ceramic-housed element bonded directly to an aluminum block [3]. Qualification: every unit went hard open within 22 cycles from -110 to +110 C using either Stycast 2850/24LV or Hysol 9309.3NA, by the ceramic housing cracking and breaking the embedded platinum wire under expansion mismatch with the substrate and the adhesive, against a requirement of 270 cycles from -105 to +40 C [2][3]. A second unit of the same type survived to cycle 585 under the MSL profiles before its resistance went from about 9000 ohms to over 1 megohm by cycle 609 [1]. Flight: six of these failed on Opportunity within six sols of landing at Meridiani Planum and 16 had failed by 2013, including all four Mini-TES external calibration target sensors across both rovers [3]. The source’s abstract instead places the failures prior to and within one sol of landing, a conflict it does not resolve.
• HRTS-5760-B-U-0-12, Honeywell. Ratings: 1000 ohms at 0 C with a coefficient of 0.00375 plus or minus 0.000029 ohm per ohm per C, 4.8 by 9.0 by 3 mm, 1.5 g, and a maximum operating current of 2 mA for a self-heating error of 1 C [3]. Qualification: 78 units bonded to seven coupons across six substrate materials with four adhesives survived 2010 cycles in dry nitrogen, with no resistance deviation, no visual change and no bond degradation on manual pull [2][3]. Flight lot screening was a further 25 air-to-air shock cycles from -55 to +125 C across about 900 MSL parts. The 2010 cycles are a simulation of three Martian years rather than an observation of them, and the two sources split them differently: 600 winter cycles from -130 to +15 C plus 1410 summer cycles from -105 to +40 C at 15 minute dwells in one, 599 winter plus 1384 summer plus four protoflight and planetary protection cycles in the other [2][3]. Flight: more than 350 units flew on Curiosity through launch, a nine month cruise, entry descent and landing and 300 sols on the surface at lows near -90 C with no failures [3]. Unbonded device-only cycling for the SMAP construction-change part reached 520 cycles from -135 to +125 C, a 260 C delta, on 20 units from three date codes, and was still running when the source was written, so that part has no bonded-configuration qualification at all. One part failed in roughly 2250 processed [3], its platinum foil cut in fabrication, although the same paper’s abstract says two failed in fabrication and its own counts reconstruct only about 1150 parts. The campaign included no pre- or post-calibration and not even an ice point measurement, so it establishes survival and never accuracy after cycling.
• FB500 flip-chip test board, manufacturer not named. Ratings: a daisy-chained flip-chip interconnect coupon carrying ten chains [2]. Qualification: three of ten chains opened by 322 cycles, opening during the cold half of the cycle and recovering hot, while the smaller FB250 coupon showed no failures after 481 cycles and repeated the result at 311 [2]. The failures were intermittent and temperature dependent rather than permanent opens, which a room-temperature continuity check would not have found.
• NVMCAM flash memory slice, JPL. Ratings: the flash memory board of the rover compute element, in a flight-like slice rather than a flight unit [4]. Qualification: solder joint cracking was observed at 500 cycles of 80 C delta, one Mars 2020 thermal cycle life, made up of 350 cycles from 0 to +80 C plus 150 from -40 to +40 C; no sample size, lot or date code is reported. Flight: the equivalent Curiosity boards accumulated 144.21 equivalent 80 C cycles on RCE-A and 429.42 on RCE-B through about sol 2320, by rainflow counting of flight thermometer data converted with a Coffin-Manson exponent of 2 [4]. The RCE-B series is a projection from the RAD750 and Base sensors rather than a measurement, at 2.38 percent average and 0.67 C RMS error against the sparse points that exist.
• PIXL Low Voltage Control Module, JPL. Ratings: an instrument electronics board qualified by cycle equivalence rather than to its own limits [7]. Qualification: 3 qualification plus 147 operational plus 212 non-operational cycles from -70 to +70 C, converted to three times life by Coffin-Manson equivalence with an exponent of 2. That range failed to envelope the matured cold non-operational protoflight temperature of -95 C, and a wet tantalum capacitor casing cracked in later assembly-level thermal testing; the project tightened the protoflight limits to -56 to +70 C operational and -60 to +70 C non-operational by using more survival heater power, rather than requalifying the board packaging to -95 C [7]. The paper’s Table 3 footnote defines its acceleration factor as the test delta over the predicted delta raised to the exponent, which is the inverse of every factor printed in the table; the tabulated factors and equivalent cycles are internally consistent and the printed formula is not, so the values above are the table’s.
• Motor encoder chip-on-board, JPL. Ratings: signal conditioning packaged inside each of 28 actuator encoder assemblies because the run to the motor controller reaches 9 m and shares cabling with motor current [9]. Qualification: 600 cycles from -130 to +60 C for the winter case and 1400 cycles from -105 to +85 C for the summer case, both 190 C ranges [9]. The result is reported only as no visible damage: no electrical parameter drift, no wirebond pull test, no cross-section and no sample size, so the cycle counts carry no failure criterion beyond visual inspection.
• HTCC alumina, co-fired platinum, NASA Glenn. Ratings: a lidless high-temperature co-fired ceramic package holding silicon carbide integrated circuits, fully exposed to the test atmosphere [11]. Qualification: both circuits operated through 60 Earth days at 460 plus or minus 5 C and 92 bar in a simulated Venus atmosphere [11]. Package insulation resistance on one of the two assemblies fell from 384 MOhm to 448 kOhm within 48 hours in chemically inert 90 bar nitrogen, before any corrosive gas was admitted, which the authors call unanticipated; the measured resistance is the package in parallel with its feedthrough wiring and the two cannot be separated. The full campaign is on NASA Glenn Extreme Environment Rig.

What the board sees mechanically

The strain limits below are a JPL review of published bend, shock and drop testing [5]; the last row is a finite element study of a flexible chassis [10]. The review performed no test of its own, names no facility, and its test vehicles are commercial and portable-electronics assemblies cycled over 0 to 100 C or -40 to +125 C, so none of the rows is at a planetary cold extreme, in vacuum, or combined with radiation.

Assembly	Measured level	Failure threshold reached
FCBGA 1521 I/O, SAC305	860 microstrain	Intermetallic fracture
FCBGA 1521 I/O, SAC305	1250 microstrain	Pad cratering
CBGA 2400 I/O, SAC387	~3000 microstrain	First failure in bend
PBGA 1517 I/O, SAC305	~6000 microstrain	First failure in bend
CGA 400	3331 microstrain	Board strain at 3000 g shock
CGA 400	4.2 drops	Mean life at 50000 g
PUFFER rigid-flex chassis	+125.38 percent stress	Non-Gaussian against Gaussian

The 1521 I/O flip-chip BGA sat on an 8-layer 93-mil board at 1 mm pitch, read by acoustic emission during monotonic four-point bend to IPC/JEDEC 9702; brittle fracture of the intermetallic preceded pad cratering on the same specimen [5]. The two larger arrays were bent on the same 6-layer flexural board at a nominal 5000 microstrain per second: the 2400 I/O ceramic BGA first failed at 930 N and reached complete failure at 1303 N and about 5000 microstrain, while the 1517 I/O plastic BGA first failed at 1187 N and reached complete failure at 1498 N and about 7000 microstrain [5]. A 1 mm board carries roughly 300 microstrain per millimeter of deflection, so a 2 mm deflection is about 600 microstrain. The specification these strains are measured against covers 1000 to 30000 microstrain per second [5], one to two orders of magnitude below the rates a shock or a drop imposes, so it does not bound the launch case.

Shock is where the column grid array data sits. A CGA 400 assembly read 3331 microstrain at the board bottom center under a 3000 g peak pulse in a dual mass shock amplifier, rising to 7251 microstrain at 50000 g, and its mean life fell from 7.2 drops at 30000 g to 5.4 at 40000 g and 4.2 at 50000 g, with two assemblies surviving 120 drops at 3000 g without failure [5]. Flight programs specify an order of magnitude below those drop levels and still miss them. The Mars 2020 rover’s avionics mounting plate carried an inherited 2000 g pyroshock specification that development mobility releases in 2019 showed was too low, and the revised specification arrived too late to requalify, so the hardware in the enclosure flew under waiver [7]. The Helicopter Base Station Electronics saw 4200 g indicated against its specified 2000 g on the same development test model, and the unit that failed was traced to a reworked integrated circuit with poor solder flow rather than to the shock level [7]. The source is a selection of highlights written before launch by the project’s own requirements engineers, and its own text retracts one of those accelerometer sets as untrustworthy because instruments of mixed size and sensitivity had been placed over structural pockets.

Alloy choice changes the bend answer and does not change it consistently. A 256 I/O plastic BGA at 1.27 mm pitch on a 1 mm board lasted 9400 cycles in four-point bend on tin-lead against 5600 on SAC405 at 60 N loading, a factor of 1.66, and at 25 N the ordering reverses to 220000 against 240000 [5]. Temperature changes it again: the strain-energy fatigue exponent fitted for 48-pin VQFN joints is 1.92 at 125 C against 0.935 at 25 C, so a room-temperature bend fatigue curve does not extrapolate to hot operation. Where the assembly itself is flexible the standard qualification margin can fail outright. A finite element study of the PUFFER rigid-flex printed circuit board chassis, whose panels are hinged with Nomex, found panel von Mises stresses up to 125.38 percent higher under a leptokurtic input of kurtosis 7.11 than under a Gaussian input of kurtosis 2.94 at a matched 13.9 grms, and a plus 3 dB envelope on the Gaussian spectrum did not bound it [10]. That work is entirely simulation with no shaker test of either chassis, reports 8 percent run to run scatter on the PUFFER case, and applies no failure criterion, only peak stress ratios.

Underfill

Underfill is the standard answer to array-package fatigue and the corpus records it failing as often as working. A flip-chip die on copper pillars, 0.04 mm bumps at 0.1 mm pitch on FR4, failed at 75 cycles from 0 to +125 C with 5 minute ramps and 10 minute dwells with no underfill; the best underfill in the same study, the one with the highest glass transition temperature, reached 1050 cycles, a factor of about 20 against a modeled prediction of 10 to 100 [6]. Expansion matching decides the direction of the effect. On a 256 I/O ceramic BGA at 1.27 mm pitch, high-lead balls on eutectic tin-lead cycled from 0 to 100 C, the non-underfilled control gave 1980 cycles to first failure and 2490 to 63 percent failure; an underfill at 75 ppm per K gave 931; and an underfill at 26 ppm per K, nearly matching the solder alloy, had not failed at 5633 cycles when every other leg including the control had [6]. A preformed underfilm on a 338 I/O molded array package at 0.5 mm pitch survived 3000 cycles from 0 to 100 C and 100 JEDEC drops at 1500 g for 0.5 ms, and failures had to be forced by raising the drop to 3500 g, with no thermal cycling penalty [5][6].

None of that is a NASA measurement. The underfill review states that test-verified guidance for NASA use of underfill is extremely limited [6], and the drop specification behind the underfilm result is written for portable products and is stated to be useful only below 15 mm of package size, which excludes the large column grid arrays space projects actually fly [5]. The modeling record is worse than the test record: the same review’s model overpredicted the non-underfilled ceramic BGA by a factor of about 2.7 [6].

Flight against qualification

Two of the entries above have a flight record to set against their chamber result, and the two records point opposite ways. The Curiosity flash memory boards accumulated 144.21 and 429.42 equivalent 80 C cycles in surface operations, against a flight-like slice that cracked at 500 such cycles, and the boards failed to mount flash on sols 200, 2172 and 2320 [4]. The chain from cycles to cracks to a failed mount remains a theory: no failure was reproduced on the ground, no cracked flight joint was ever inspected, the sol 2172 root cause was lost when the diagnostic data was compacted, and the sol 2320 investigation was open at publication. The counting also rests on an assumed Coffin-Manson exponent of 2 [4], which linearizes the damage from the many small diurnal cycles, whose delta T is about 15 C with the population concentrated between 10 and 20 C, against the few large ones.

The temperature sensors went the other way. The type X element had already gone hard open at 22 chamber cycles in the configuration that flew [2], and it then failed on both Mars Exploration Rovers within sols of landing and kept failing for years [3]. Its replacement survived 2010 bonded cycles in qualification, and more than 350 units then flew on Curiosity with no in-flight failure [2][3]. What separates the two is a bonding process and a ceramic housing rather than a sensor technology, and the chamber identified the failing configuration before flight confirmed it.

Standards fix the shape of these campaigns without fixing their outcome. Mars 2020 surface hardware was cycled 3015 times, three times the 1005 mission, ground and cruise cycles, as 900 winter cycles from -115 to +15 C plus 2115 summer cycles from -80 to +50 C at a maximum 130 C delta and a ramp of at most 5 C per minute, and no hardware of any class was permitted to be tested below -135 C [7]. The motor qualification campaign for the same rover ran the same 3015 cycles to represent 4.5 Mars years, but prints the seasonal bands as -80 to +85 C for summer and -115 to +50 C for winter [16], which neither source reconciles against the other. The European equivalent asks for 4 temperature cycles at equipment level, a rate of change below 20 K per minute and a dwell of at least 2 hours before a functional test, at a chamber pressure of 1e-5 hPa or better [15]. Its input tolerance is defined only above 80 K, and its qualification margin of plus and minus 5 C over the acceptance extremes may be reduced below -170 C, so the coldest destinations carry the least margin under the standard that governs them [15].

Coatings, plating and staking

Conformal coating is carried for several unrelated reasons at once, and each program qualifies it against the one that constrained its design. The Mars 2020 motor boards were coated in Parylene HT to protect against ground storage corrosion, to protect against short circuits from debris in weightless conditions, and to suppress tin whisker growth [16]. The SHERLOC laser power supply needed a fourth job, corona suppression in the Martian atmosphere: its original stack of five layers of CV1144 plus one of CV1142 allowed three arcs in the qualification unit and lost the pulse-width-modulation chip, and the redesigned stack of five CV1144, one CV1142, one CV1144 and one thick CV1442 layer holds a minimum 100 mils and comes within 10 mils of the cover [7]. Encoder wirebonds went the other way and had their coating removed: the MER epoxy overcoat was deleted on Curiosity so the chip-on-board assembly would survive thermal cycling, leaving the bonds exposed under parylene alone, which is also the board moisture seal, with handling protection carried by test fixtures, assembly tooling and a per-unit shipping fixture [9].

Plating is normative rather than measured. Tin in mission-critical hardware must carry at least 3 percent lead by weight or another proven alloying element, or else comply with GEIA-STD-0005-1A and Control Level 2C of GEIA-STD-0005-2A, and gold must be removed from at least 95 percent of any surface to be soldered to prevent embrittlement of the joint [8]. Nonmetallic materials pass at not more than 0.1 percent collected volatile condensable materials and 1.0 percent total mass loss by ASTM E595-15, with hardware containing a failing or unidentified material baked at the maximum temperature the component tolerates or 10 C above its maximum predicted operating temperature. These are pass criteria and never results, every one of them can be departed from through a Materials Usage Agreement, and the 1.0 percent limit carries a written escape for a higher mass loss with no effect on functionality, so a compliance statement is not evidence that a material met the limit [8].

Staking was the only degradation the temperature sensor campaign recorded. Across 78 units on six substrate materials with four adhesives run to 2010 cycles, what was lost was cable staking adhesion and nothing else [2].

The package inside a radiation test

A package is part of the specimen in a heavy-ion beam, and three results in the corpus turn on that. A 256 K SRAM tested in its plastic package returned an apparent single event upset cross section 100 times lower than the same part delidded, because the ions of a degraded krypton beam ranged out in the package before reaching the sensitive volume; the plastic result is an artifact of the packaging, not a hardness [13]. Aluminum package material has an ion range of 0.85 relative to silicon, which is the conversion that decides whether a beam reaches a die at all [14]. Package chemistry can also change the outcome without changing the ion: the ELDRS-immune LM139A comparator’s delta input bias current rose by an order of magnitude at 10 rad(Si) per second, in a campaign the compendium compiles rather than performed, after its package had been soaked at 100 percent and at 1 percent molecular hydrogen content, against unsoaked high and low dose rate baselines, and its input offset voltage rose with it [12]. Package hydrogen content is a variable that no dose rate qualification detects.

The part-level radiation results for the comparators and operational amplifiers those findings come from are on Power and Batteries, and the beam facilities are under Testing.