At final test, every packaged chip is measured against its datasheet and sorted by what the tester finds, good, bad, or which grade it makes. Because a chip’s parameters drift with temperature, and the datasheet promises behaviour across a wide range, the chip is tested hot, cold, and at room temperature. The handler and chamber together bring each chip to temperature, hold it there in contact with the test socket while the tester measures it, and do this thousands of times an hour. This is final test, a short measurement at temperature, a sort and never a screen, where a single bad contact can fail a good chip.
A chip that has been fabricated, packaged, and is ready to ship still has to be proven, one part at a time, against everything its datasheet claims. That proving is final test, the last gate before a chip is sold. A tester, an expensive instrument, applies signals to the chip and measures what comes back, checking that every parameter falls inside the limits the datasheet sets. A handler is the machine that presents each chip to the tester, takes it away when the test is done, and sorts it by the result. Between them sits the reason a chamber is part of the cell at all. A datasheet does not promise behaviour at one comfortable temperature. It promises behaviour across a range, from well below freezing to well above the boiling point of water. A chip’s parameters shift as its temperature changes. To prove the datasheet, the chip has to be tested at room temperature and at the hot and cold extremes as well, which means the handler has to bring each chip to temperature and hold it there while the tester does its work. The integration of handler, tester, and chamber is what makes that possible at the speed a production line demands.
Final test measures a finished chip against its specification and sorts it by what it finds. The tester runs the chip through its parameters, the voltages it must hold, the currents it must draw, the speed it must reach, the functions it must perform, and compares each against the datasheet limit. A chip that meets every limit passes. A chip that misses one fails. Many chips are graded, sorted into bins by how well they perform, the faster parts marked for one price and the slower for another, the same design sold at different grades by what each unit can do. The handler is the machine that makes this happen at volume. It takes chips from trays or tubes, brings each to the tester, holds it while the tester measures, and drops it into the bin its result calls for. What final test is not is a screen. A screen, like a burn-in or a stress screen, stresses every unit to drive out the ones that would fail early, and removes the weak from the population. Final test does something different. It measures every unit and classifies it, sorting the good from the bad and the fast from the slow. The chamber’s part in it is to put each chip at the temperature the measurement calls for, because the answer to whether a chip meets its datasheet depends on how hot or cold the chip is when the question is asked.
A datasheet is a promise made across a range of temperatures. A chip keeps that promise differently at each. Silicon changes its behaviour as it heats and cools. Leakage currents climb steeply as a chip warms, so a part that draws a tiny standby current at room temperature can draw far more when hot, missing at the top of its range a limit it met cool. Speed shifts the other way for many parts, a circuit timing that is comfortable warm tightening as the part goes cold, so a chip can fail a timing test at the low end that it passed at room. Thresholds move, offsets drift, the entire web of parameters slides as the temperature changes. A chip proven only at room temperature is proven only at room temperature, which is the one condition a customer is least likely to hold it at. Final test is therefore done at temperature, and usually at more than one, hot to catch the parameters that fail when warm and cold to catch the ones that fail when chilled, with room temperature between them. This is tri-temperature testing. The range it covers is wide, often from around fifty-five degrees below zero to well over a hundred and fifty above. The hard part is doing it fast. Each chip has to be brought to an accurate temperature, held there steady while a measurement that may take only a fraction of a second is made, and moved on, with the next chip already on its way to the same temperature. Bringing a small part to a precise temperature quickly, holding it there through the test, and never making the costly tester wait, is the problem the handler and chamber exist to solve. The cost of getting it wrong runs both ways. A part shipped because it passed at room temperature, then failing in a customer’s hot or cold use, is a field failure the test was meant to prevent, the worst outcome of all. A part rejected because the test took it to a temperature its datasheet never claimed is a sound part thrown away, a yield loss from testing too hard. The temperatures the test uses are not arbitrary. They are the limits the datasheet names, the hot and cold corners the part is guaranteed to meet, tested at the edges of what was promised, so that a pass means the promise holds and a fail means it does not. Choosing those temperatures, and holding each chip at them accurately, is much of what makes the test mean what it says.
A chip proven only at room temperature is proven only at room temperature.
Handlers come in a few mechanical forms, each suited to a kind of part and a speed. A gravity handler lets chips slide down a track under their own weight, through a test site, and into bins below, a simple and quick arrangement that suits packages with leads that travel well down a chute. A pick-and-place handler uses a head that lifts each chip, moves it to the contactor, and places it down, a flexible form, able to handle the delicate packages and the ball-grid arrays that cannot be slid down a track, and able to test many in parallel by carrying several at once. A turret handler mounts chips around the rim of a rotating wheel that indexes them past stations in turn, reaching the highest speeds for small parts in volume. The choice among them follows the package and the throughput the line needs, the gravity handler for simple parts at speed, the pick-and-place for awkward packages and parallel testing, the turret for small parts in the highest volumes. What they share is the task at the centre, bringing each chip to a contactor at the right temperature so the tester can reach it. Parallelism cuts across all three forms. Because a tester can often measure several chips at once, a handler that presents many parts to the contactor together multiplies the throughput, testing eight or sixteen or more in the time one would take. The pick-and-place form lends itself to this, carrying a row of chips to a multi-site contactor in one motion, which is part of why it dominates for the complex packages and high site counts of modern parts. The mechanical form and the degree of parallelism are chosen together, to match the package the line runs and the rate it has to hold.
Getting a chip to temperature can be done two ways. The difference matters chiefly for parts that make their own heat. The simpler way is a thermal soak. The handler holds the chip in a conditioned space, hot or cold, long enough for it to reach the temperature throughout, then tests it while it sits at that soaked temperature. A soak works when the chip’s own heat during the brief test is small enough to ignore. The harder case is a chip that dissipates real power while it is tested, a processor or a power device that heats itself the moment the tester drives it, so that a part soaked to a target temperature climbs above it the instant it is powered, and is no longer at the temperature the test was meant to use. For these, a soak is not enough. The cell uses active thermal control. A sensor close to the chip, often built into the contactor itself, reads the device’s temperature in real time, while a fast heater and cooler at the contact drive that temperature to the target and hold it there against the chip’s own self-heating, closing the loop on the device’s junction temperature where a soak settles for the air around it. Active thermal control is what lets a high-power chip be tested at a true, held temperature while it burns watts into the socket. It is one of the harder things a test cell is asked to do.
The point where the chip meets the tester is the contactor, through which everything the test measures passes. The contactor is a socket that presses against the chip’s leads or balls, making a temporary electrical connection between each one and the tester’s measuring circuits. It has to make that connection cleanly for every contact, dozens or hundreds of them on a modern package, in the moment the chip is pressed down, and break it cleanly when the chip is lifted, thousands of times an hour without wearing out or drifting. The contact it makes has to be good, a low and stable resistance on every pin, because the tester’s measurement is only as true as the connection carrying it. The contactor is also where the mechanical world of the handler meets the electrical world of the tester, the one place a misplacement of a fraction of a millimetre, a speck of debris, or a worn contact turns into a measurement error. For all that it is a small passive part, the contactor is among the hardest-working pieces of the cell, because the entire result rests on the quality of the contact it makes in a fraction of a second, at temperature, over and over.
The contactor’s importance shows sharpest in how it fails. When a contact does not connect, the tester cannot reach the pin behind it, and reads the chip as failing a test the chip would have passed with a clean connection. The chip is sorted into the fail bin, scrapped or set aside, although nothing is wrong with it. This is a false reject, a good chip lost to a bad contact where no real defect exists. It is one of the costlier ways a test cell goes wrong, because it throws away parts that were fit to sell. False rejects from contact trouble can masquerade as a drop in yield, sending engineers hunting for a fabrication problem that is not there, when the fault is a worn socket or a speck of debris on a pin. A well-run cell guards against this, monitoring contact quality, cleaning and replacing contactors on a schedule, and retesting a suspected false reject to tell a true fail from a contact fail. The distinction matters in money, since every good chip thrown away as a false reject is yield lost for no reason. At the volumes a handler runs, even a small rate of false rejects is a real cost. The contactor that makes a clean connection every time is doing one of the more valuable jobs in the cell. The opposite error is rarer at final test but graver, a bad chip passed as good, an escape that reaches a customer. A contact fault tends toward the false reject, reading a good part as bad, since an open contact looks like a failure. An escape needs the measurement itself to be wrong in the chip’s favour, which a clean test guards against. The cell is tuned to keep both rates low. The false reject from a bad contact is the everyday loss it watches hardest, because it happens unseen, part by part, showing only as a yield lower than it should be.
Testing a chip cold raises a problem the hot end never does. When a chip and its socket are taken below the dew point of the surrounding air, moisture condenses on them, beading the cold surfaces with water and, colder still, with frost. On a contactor, that moisture is a direct threat to the contact, since water between a pin and a pad spoils the connection the test depends on. Frost, colder still, can hold a contact open. The cold test environment therefore has to be kept dry, purged with a dry gas or sealed and held below the dew point so that no moisture can reach the cold parts. Managing condensation is a standing requirement of cold testing, one the cell solves by controlling the humidity around the cold zone so that no room air, with its moisture, reaches a surface cold enough to condense it. The same physics of a cold surface meeting damp air appears wherever parts move between temperatures, a broader subject handled in its own right. At the contactor, the point is narrow and sharp, a cold contact must be kept dry, because a drop of condensation there is indistinguishable, to the tester, from a defect in the chip.
The tester is the costliest thing in the cell, and every second it sits idle is money lost. The handler is judged above all on how well it keeps the tester fed. A tester measuring a chip in a fraction of a second can be starved by a handler that takes seconds to bring the next chip to temperature and into contact, leaving the expensive instrument waiting on the mechanics. The handler’s index time, the time to remove a tested chip, place the next, settle it to temperature, and make contact, is what sets the throughput of the entire cell, counted in units per hour. Temperature is often the slowest part of that cycle, since bringing each chip to a soaked temperature takes time the tester does not need, which is why a tri-temperature test, with its waits for the part to settle at each temperature, can be far slower than a test at room temperature alone. The cell fights this with parallelism, testing many chips at once so the tester works on a batch while the handler conditions the next, and with thermal designs that bring parts to temperature quickly. The entire arrangement is built around keeping the tester measuring as much of the time as it can, because the tester’s time is the expensive resource the rest of the cell is there to use well.
When the tester finishes a chip, it returns a bin number, a verdict that says more than pass or fail, naming which category the chip belongs to. The handler reads that number and sorts the chip into the matching physical bin, the passing parts by grade, the failures by kind. Grading is where final test gains much of its value, because the same design often yields parts of different capability. Sorting them by what each can do lets the faster parts be sold as a faster, dearer product and the slower as a cheaper one, the test turning a spread of performance into a set of graded products. The result for each chip is also recorded, tied to the chip’s identity, so the outcome of its test joins the data the factory keeps on every part it makes. How that record is gathered and tied into the factory’s systems is the work of the line’s software, handled in its own right. What the handler contributes is the physical act of sorting, putting each chip where its verdict sends it, and the data point that says what that verdict was, for the one part now dropping into its bin.
Testing at three temperatures can be arranged two ways. The choice is a trade between speed and complexity. The straightforward way is three passes, running the entire lot through the handler at one temperature, then again at the next, then at the third, so each chip is tested three times, once at each temperature. This is simple to build and slow to run, since the lot waits through three full runs and the cell settles to a new temperature between them. The faster way puts more than one temperature in a single pass, a handler with separate hot, cold, and room zones through which each chip travels in turn, tested at each as it goes, so a chip is brought to all three temperatures and measured at each in one trip through the machine. A single pass at three temperatures is quicker, since no chip is handled three times and the temperatures run in parallel across the machine. It asks more of the handler, which now has to hold and move parts through several temperature zones at once. The choice between them is the usual one of throughput against the cost and complexity of the machine, decided by the volume and the value of the parts being tested. Not every part needs all three temperatures either. A part whose parameters are known to fail first at the hot corner may be tested hot and at room, skipping cold, where the data show cold adds nothing. The number of temperatures, and which ones, is set from what is known about how the part drifts, testing at the corners that catch real failures and sparing the cell the time of a temperature that would catch none. Tri-temperature is the full case. Many parts are tested at fewer, by a judgement of where their datasheet is at risk.
The clearest way to place final test among the other things a chamber does for electronics is to see it as a sort. A burn-in and a stress screen take a population and remove its weak members, stressing every unit to make the early failures show and pulling them, so what ships has survived a hard window. Final test takes a population and classifies it, measuring every unit against its datasheet and routing each to where its result belongs, the good to shipment, the failed to scrap, the graded to their grades. The screen asks whether a unit is weak. The sort asks what a unit is, and where it belongs. The chamber serves both, but serves them differently, holding a unit at a stress for a screen and at a measurement temperature for a sort. For final test, the chamber’s gift is precision, an accurate, held temperature at which the datasheet can be proven, delivered for every chip at the speed a line demands. The handler, the tester, and the chamber together turn a stream of finished chips into sorted, graded, accounted-for parts, each one measured at the temperatures its datasheet promises and sent to where its measured worth belongs.
Because a datasheet promises behaviour across a wide temperature range, and a chip’s parameters drift as it heats and cools. Leakage rises steeply when a chip is hot, and timing can tighten when it is cold, so a part that meets a limit at room temperature may miss it at an extreme. Testing hot, cold, and at room, called tri-temperature testing, proves the chip across the range a customer may use it in.
A soak brings the chip to temperature in a conditioned space and tests it as it sits there, which works when the chip makes little heat of its own during the test. Active thermal control adds a sensor near the chip and a fast heater and cooler at the contact, closing the loop on the device’s own temperature. It is needed for high-power parts that heat themselves the moment they are powered, since a soaked part would climb off its target temperature without it.
A false reject is a good chip sorted into the fail bin by a bad contact, with no real defect in it. If a contactor pin does not connect, the tester cannot reach that pin and reads the chip as failing, with nothing wrong in the silicon. False rejects from worn or dirty contactors can look like a drop in yield, so a well-run cell monitors contact quality, cleans and replaces contactors, and retests suspect parts to tell a true fail from a contact fail.
A screen, such as burn-in or environmental stress screening, stresses every unit to drive out the ones that would fail early, removing the weak from the population. Final test instead measures every unit against its datasheet and sorts it, judging good from bad and grading the passing parts by performance. A screen asks whether a unit is weak; final test asks what a unit is and where it should go. It is a sort. It is not a screen.
The tester is the costliest part of the cell, so time it spends waiting is money wasted. A handler that is slow to bring the next chip to temperature and into contact starves the tester, which can measure a chip in a fraction of a second. The handler’s index time and the temperature settling, slowest at the cold and hot extremes, set the throughput. Cells fight this with parallel testing and fast thermal designs, to keep the tester measuring as much of the time as possible.
Envsin robotic temperature chambers for IC final test handler integration, tri-temperature sorting and reliability testing.