AEC-Q104 decides whether a multichip module is fit for a car. A module packs several chips into one body. That body carries many interfaces, every one a place that can fail. Thermal shock is the stress that finds the weakest of them. It slams the module between a cold extreme and a hot one in seconds. The part that cannot take the slam gives way first. AEC-Q104 goes further than the chip standards before it. It tests the module after it is soldered to a board, so the joints that hold it down are judged too.
AEC-Q104 is the stress-qualification standard the Automotive Electronics Council wrote for multichip modules. A multichip module gathers several die into one package. It might be a processor beside its memory. It might be a sensor with its readout circuit. It might be a stack of chips built upward to save room on the board. The standard treats the assembly as the thing being qualified, not the individual chips inside it. It pulls the module through a battery of stresses to prove it will survive the life of a car. The stress that does the sharpest work on a module is thermal shock, the fast slam between a cold extreme and a hot one. A single chip can take that slam. A module made of many chips, joined by many interfaces, gives the slam far more places to find a weakness.
The reason a module is harder to qualify than a chip is the number of joints inside it. A single integrated circuit has its die, its attach to a lead frame or a substrate, and its bonds to the outside. A module multiplies all of that. Each die in the module is attached to a common substrate, by solder bumps or by fine wires. Each of those attachments is an interface. Some die sit beside others and connect to them directly, die to die, across another set of joints. The module itself then sits on a package substrate. That substrate connects down to the circuit board through an array of solder balls. Every one of those levels joins two materials that expand by different amounts when the temperature changes.
That difference in expansion is the root of the trouble. Silicon expands little with heat. The organic substrate of a package expands several times as much. Copper, solder and the moulding compound each move by their own amount. In numbers, silicon expands by about three parts per million for each degree of temperature. An organic substrate expands by fifteen to twenty, and solder and copper by more again. A swing of two hundred degrees turns those small per-degree differences into real movement at every joint. When the temperature swings, the materials pull against one another at every interface. The strain gathers where two unlike materials meet. A module has more of those meeting places than any single chip can. A heterogeneous module, built from die of different sizes and different functions, holds the widest spread of mismatch of all. Thermal shock is built to load every one of those interfaces at once. It lets the weakest among them declare itself.
Thermal shock is the harshest of the temperature tests because of how fast it moves. The module is taken from a cold extreme to a hot one, or back, in seconds, where a cycle would take minutes. The surface of the module reaches the new temperature almost at once. The inside lags behind, still near the old one. For a brief moment the module holds a steep temperature gradient through its own thickness. That gradient is what does the damage. The detailed physics of how a fast transfer builds a transient gradient belong to the thermal-shock method itself. What matters for a module is the breadth of the target. The gradient lands across many interfaces at the same time, each already strained by the mismatch in expansion. The steepness of the shock multiplies that strain far past what a gentle change would apply. A gentle temperature cycle gives each interface time to settle into the new temperature before the strain peaks. A shock denies that time. The strain arrives while the gradient is steepest, when the surface and the core are furthest apart, so each interface meets its worst load and its fastest change at the same instant. That combination, the largest strain delivered the fastest, is what tests a many-interface module hardest.
The interfaces fail in ways a module engineer knows by name. The die-attach under a chip can delaminate. It peels from the substrate at a corner and spreads inward, lifting heat off its escape path and weakening the bond. The underfill that grips a flip-chip die can crack, freeing the solder bumps to fatigue. A solder bump can crack outright, opening a connection between a die and the substrate. A wire bond can lift at its heel. The solder balls under the package can fatigue against the board. A module offers the shock all of these targets in one part. The test is built so that the first one to give does so on the bench, where it is caught, long before it would surface in a car.
Heterogeneous modules feel this hardest. When a large logic die sits beside a small memory die, the two warp by different amounts and at different rates as the shock hits. The substrate between them has to absorb the difference. The wider the spread in die size and material, the more uneven the strain. A wider spread gives the shock more to break. This is the reason the module standard leans on thermal shock so heavily, where a simpler part might lean on gentle cycling alone.
A chip has a few interfaces. A module has many, and the shock tests them all at once.
Temperature cycling and thermal shock sound alike. They are not. Both swing a part between a cold extreme and a hot one. The difference is the speed of the transition. The speed changes what the test finds. Temperature cycling moves slowly enough that the entire part reaches each temperature together. Its stress comes from the materials expanding by different amounts at one shared temperature. Thermal shock moves so fast that the part cannot keep up with itself. Its stress comes from the surface and the core sitting at different temperatures in the same moment. A module can pass a gentle cycle and still fail a shock, because the shock adds the transient gradient on top of the mismatch. AEC-Q104 calls for the fast transition to apply that harder load. It ties the conditions to the established JEDEC method, so a result can be compared across makers.
A thermal shock chamber is built around the speed the test demands. It holds two separate zones. One is kept at the hot extreme, the other at the cold, each ready and waiting at its temperature. The parts ride a basket that moves between the zones. The point of the design is the transfer. The basket carries the modules from one zone into the other in a few seconds, so the parts meet the new extreme almost without a ramp. The transfer time is a defining specification of the test, not an afterthought. A slow transfer would soften the shock into something closer to a cycle. The harder stress the test exists to apply would be lost.
The zones have to recover fast as well as hold steady. When a cold basket full of modules drops into the hot zone, it pulls the zone’s temperature down. The zone has to drive itself back to the extreme quickly, so the parts reach the full temperature for the dwell that follows. A chamber that cannot recover under a heavy load is testing a milder shock than it claims. The cold side carries the same demand in reverse, with refrigeration that holds its grip when a hot load arrives. The size of the load sets how hard this is. A basket carrying many modules holds real thermal mass. Dropping that mass into a zone is a large, sudden demand on the zone’s heating or its cooling. A chamber sized for the load drives both zones with the power to swallow the basket and steady itself before the dwell is counted. A shock chamber proves itself by keeping both zones at their extremes and moving the parts between them faster than the parts can follow.
There are two ways to build the slam. They differ in how hard they hit. An air-to-air chamber holds two air zones and shuttles the basket between them. Air carries heat slowly, so an air shock is the gentler of the two, though it is still far faster than a cycle. A liquid-to-liquid chamber drops the parts from a hot bath into a cold one, using inert fluids that carry heat far faster than air. The fluid shock is the harder hit. It reaches the surface of the part almost on contact. It is reserved for the strictest qualifications. The choice of medium sets the severity. The standard names which one a module must face. Either way, the chamber owes the test the same thing: deliver the transition the method specifies, to every part, for every cycle of the run.
A shock test is written as a profile. The chamber has to hold every term of it. The extremes are set, often a cold soak near minus fifty-five degrees and a hot soak up toward a hundred and twenty-five or a hundred and fifty, matched to the grade the module claims. The dwell at each extreme is set long enough for the entire module, surface and core, to reach the temperature before the next transfer. The transfer window is set short, the few seconds that keep the shock a shock. The number of cycles is set high, hundreds or more, to reach the fatigue life a car would impose. A module passes when it survives the full count with its interfaces intact and its function sound. The chamber has to deliver each of those terms without drift. A dwell cut short, or a transfer stretched long, changes the stress the module saw.
The hardest modules to qualify are the ones built upward. A stacked module places die on top of die, joined through their thickness, to fit more function into the same footprint. Stacking adds interfaces a flat module does not have. Thin die are bonded together and linked by connections that run vertically through the silicon. A thermal shock now strains those vertical joints along with the lateral ones. Those vertical connections, run through the silicon itself, are short and stiff, with little room to flex as the stack expands and contracts. A thin die in the middle of a stack is gripped above and below. A shock that warps it has nowhere to send the strain but into the joints at its faces. The stack also holds a temperature gradient through its height during the transfer, top to bottom, that adds to the gradient across its width. The specific physics of a wide flat advanced package, and of a tall vertical memory stack, each belong to their own subjects. What AEC-Q104 takes from them is one lesson. The more a module is built up and out, the more interfaces a shock has to work. The chamber then has to deliver the slam to every one of them with that much more care.
AEC-Q104 broke new ground by testing the module after it is on the board. Earlier chip standards qualify the part by itself, in its package, on a test fixture. A module is meant to be soldered down to a circuit board. The array of solder balls that holds it there is one more set of interfaces a thermal shock can fatigue. Board level reliability, the part of AEC-Q104 that puts a soldered module through the stress, is the first such requirement in an automotive component standard. It matters because the module and the board warp by different amounts as the temperature swings. That warping concentrates on the outermost solder balls under the package. The outermost balls crack first. The reason is leverage. The package and the board meet at a plane of solder balls. When the two warp by different amounts, the corners of that plane move the furthest relative to one another. A ball at the corner takes the largest shear at each swing, so it fatigues ahead of the balls near the centre. A module can read perfect on a fixture and still open a corner ball after a few hundred shocks on a board. A module can be sound in its package and still fail at its feet once it is mounted. The standard tests the joined assembly, so the feet are proven along with the body. The chamber has to shock the populated board the same way it shocks a bare part.
AEC-Q104 organises its work into categories, lettered A through H, that group the stresses and the checks by the failure each addresses. Some are the reliability stresses themselves: temperature cycling, the fast transition that serves as thermal shock, drop and mechanical loads, low-temperature storage, and the start-up and temperature-step routines that catch early-life faults. Others are the failure inspections that read the result. X-ray imaging looks through the package for cracked joints and voids. Acoustic microscopy listens for the echo a delamination returns, mapping where a layer has lifted. Destructive physical analysis cuts the module open to confirm what the non-destructive reads suggested. Acoustic microscopy works by sending ultrasound into the part through a water couplant. Sound passes through solid material and reflects sharply off any air gap, so a delamination or a void returns a bright echo that maps as a flaw against the sound regions around it. X-ray sees density instead, throwing the metal of a joint into shadow, so a crack or a void in a solder ball shows as a break in the shape. Between them the two reads find damage no electrical test would yet reveal. A module passes only when the stresses leave it intact and the inspections find nothing growing inside. The standard draws its parts from more than one production lot and holds them to an acceptance of zero failures, the same governance that underpins every automotive qualification.
A module carries a grade, and the grade is the temperature span it has been proven to take. The shock extremes follow the grade. A module rated for the harshest mounting points has been shocked between the widest extremes, a deep cold to a high heat. A module for a milder location is proven across a narrower span. The grade fixes how far apart the two zones sit, which sets how violent the slam is. A wider span means a steeper drop in the same transfer time. A module qualified to a wide span has survived a harder shock at every cycle. The chamber serving the top grades has to hold a colder cold and a hotter hot, and recover to both under a loaded basket.
A few hundred shocks in a chamber stand in for fifteen years of a car’s thermal life. The bridge is acceleration, the principle that a harder, faster stress ages a joint along the same path as the gentle thermal swings of service, only quicker. Each shock works a tiny step of fatigue into the weakest interface, the same fatigue a day of heating and cooling would add over years. The models that tie a cycle count to field years rest on the chamber having delivered each shock in full. A transfer that ran slow, or a dwell that fell short of the extreme, is a shock the module never fully took. A field-life claim built on softened shocks claims a durability the test did not prove. The count is set with that bargain in mind. A few hundred shocks is chosen to match the fatigue a target life would impose, with a margin folded in for safety. Run too few shocks and a weak interface survives the test only to fail in the field. Run the full count at full severity and the weak interfaces are flushed out on the bench. The number is a promise the chamber keeps only by delivering every shock the number assumes.
The reads that decide a module are part electrical and part internal. An interface that has opened shows as a broken connection, a net that no longer carries current, caught by an electrical test between the shocks or at the end. Much of the damage hides inside and grows before it opens, so the standard leans on the inspections. Acoustic microscopy finds a delamination as a bright echo where a layer has parted, long before the bond fails outright. X-ray finds a cracked solder ball or a void as a shadow in the joint. A module read clean electrically can still carry a delamination halfway across a die-attach, on its way to a failure a year out. The inspection is what catches it. The chamber’s contribution to all of this is the honesty of the shock behind it. Damage found after a clean, full-severity shock points at the part. Damage found after a softened shock points nowhere, since a transfer that wandered cannot be told from a part that was weak.
A module often sits at the heart of a system, carrying more function than any single chip around it. A processor module that fails takes an entire control unit down with it. The cost of qualifying it wrongly is steep, which is the reason the standard tests the module as an assembly and inspects it inside and out. A shock test that passed a module with a hidden delamination would let a latent fault into a car. It would surface as an intermittent failure on a cold morning years later, the hardest kind to find in the field. A true, full-severity shock, read by an honest inspection, is what keeps that latent fault on the bench. A module wins its place in a car by surviving the slam and showing nothing growing inside. This is the reason the inspections weigh as heavily as the stresses. A module that survives the shocks electrically has cleared only half the bar. The acoustic and X-ray reads have to find no delamination spreading and no solder ball cracked before the part is called good. A latent flaw that slips past the electrical test and shows under the inspection is a flaw caught in time. The standard keeps a doubtful module out of a car on the strength of an inspection alone.
What AEC-Q104 asks of a chamber is a true shock, delivered the same way every time, to a part that hides its weaknesses inside. It has to hold two zones at their extremes, a deep cold and a high heat, and keep them there as loaded baskets pull at them. It has to move the parts between the zones in the few seconds that keep the shock a shock. It has to recover each zone fast enough that the dwell reaches the full extreme. It has to do this for hundreds of cycles without drift, on a bare module and on one soldered to a board. A chamber that delivers all of that lets a maker grade a module and stand behind it, knowing the slam that proved it was the slam the standard named. The grade is a promise about the thermal life the module can take. The chamber is what makes the promise true, one honest shock at a time.
AEC-Q104 is the Automotive Electronics Council standard for multichip modules, packages that hold several die in one body and solder down to a circuit board. It qualifies the entire module as an assembly, not the chips inside it separately. It runs the module through reliability stresses such as thermal shock and temperature cycling, then inspects it by X-ray, acoustic microscopy and destructive analysis to confirm nothing has cracked or delaminated inside.
Both swing a part between a cold extreme and a hot one. Thermal shock makes the transition in seconds, where temperature cycling takes minutes. The fast transfer means the surface of the module reaches the new temperature while the core lags, so the module holds a steep gradient through itself for a moment. That transient gradient stacks on top of the usual expansion mismatch, so a module can pass a gentle cycle and still fail a shock.
A module has many more interfaces than a chip. Every die attaches to the substrate. Some die connect to one another. The package connects down to the board. Each level is a meeting of materials that expand by different amounts. A thermal shock strains all of those interfaces at the same instant and finds the weakest. A heterogeneous module, built from die of different sizes, has the widest mismatch and the longest list of targets for a shock to break.
Board level reliability tests the module after it is soldered to a circuit board. AEC-Q104 was the first automotive component standard to require it. The module and the board warp by different amounts as the temperature swings, which concentrates strain on the outermost solder balls under the package. Those balls crack first. A module can be sound in its package and still fail at its solder joints once mounted, so the standard proves the joined assembly.
It holds two zones, one at the cold extreme and one at the hot, and moves the parts between them in a few seconds, so the transition stays a shock. It has to recover each zone fast when a loaded basket pulls its temperature off the mark, so the dwell reaches the full extreme. The transfer time and the dwell are defining terms of the test. A chamber that stretches the transfer or shortens the dwell delivers a softer shock than the standard named.
Envsin reliability and environmental test chambers for automotive multichip-module qualification.