Wafer probe testing, also called wafer sort or circuit probe (CP), is the electrical test of every die on a finished silicon wafer before the wafer is sawn and the dice are packaged. A prober steps the wafer under a probe card whose fine needles touch each die's pads, and automatic test equipment (ATE) applies power and test patterns and measures the response. Each die is marked pass or fail on a wafer map, so only good dice go on to the more expensive packaging and final test steps.
Courtesy: Wikipedia
Wafer probe testing is a type of testing that is performed on individual devices on a silicon wafer to verify basic functionality and parametric performance. During wafer probe testing, electrical probes are used to make contact with individual devices on the wafer and measure various electrical characteristics, such as voltage, current and capacitance. This information is used to verify that each device is functioning as intended and meets the required specifications.
Key takeaways
- Wafer probe is the first electrical test of each complete die, done while the dice are still on the wafer.
- Its main job is to stop bad dice from being packaged, because packaging and final test cost far more per unit than probing.
- The hardware is a prober (wafer handling and alignment), a probe card (the needles) and an ATE (the instrument that runs the test program).
- Results are stored as an electronic wafer map with a bin code per die; older flows marked bad dice with ink.
- Probe is not the same as the parametric test (WAT or PCM) done on scribe-line structures, and it does not replace final test after packaging.
Where wafer probe sits in the manufacturing flow
Wafer probe testing is typically performed early in the back half of manufacturing, after front-end processing is complete and before the devices are packaged and assembled into finished products. This helps to catch defects early, reducing the cost and time spent on parts that would fail anyway. A typical sequence looks like this:
- Front-end fabrication: transistors and interconnect are built on the wafer.
- Wafer acceptance test (WAT), also called process control monitor (PCM) test: the fab measures test structures in the scribe lines (transistor threshold voltage, sheet resistance, contact resistance) to confirm the process stayed in its window. This tests the process, not the product dice.
- Wafer probe (wafer sort, CP): each product die is powered and tested. Many products run more than one probe insertion, for example one at room temperature and one hot or cold.
- Back-grind, dicing and die pick: the wafer is thinned and sawn; the pick-and-place tool reads the wafer map and picks only good dice.
- Assembly: good dice are packaged.
- Final test (FT) and, where required, burn-in and system-level test: the packaged part is tested again, often at full speed and across temperature.
For more on how these stages fit together, see the other articles in this series, for example the different kinds of silicon testing and silicon production testing.
How the hardware works
The prober
The prober is the wafer handling robot shown in the photo above. It loads a wafer from a cassette or FOUP onto a vacuum chuck, aligns it optically to the pad pattern, and steps from die to die. Each contact is called a touchdown. The chuck can usually be heated or cooled, so the same die can be tested at, for example, -40 C and +125 C for automotive parts. The prober also controls overtravel, the small extra distance the wafer is pushed up after first contact so that the probe tips scrub through the thin oxide on aluminium pads.
The probe card
The probe card is a printed circuit board carrying the probe needles, custom made for each product. Common styles are cantilever needles (older, lower pin counts), vertical probes (for area-array pads and bumps) and MEMS probe cards (lithographically made tips for very fine pitch and high pin counts). Probe cards are expensive, wear with use and need regular cleaning, so card life and maintenance are real production costs.
The ATE and the test program
The ATE supplies power, drives input patterns, measures outputs and runs the test program written by the product and test engineers. It talks to the prober over an interface so the prober knows when to step to the next die and which bin to record.
What is actually tested at probe
Wafer probe testing is useful for checking that each die works and for characterising the performance of individual devices, as it allows the measurement of key parameters such as power consumption, speed and operating temperature range. A typical digital or mixed-signal probe program includes:
| Test | What it checks | Typical failure it catches |
|---|---|---|
| Continuity / open-short | Every pad connects through its ESD diode | Bad probe contact, open or shorted pads |
| Leakage and IDDQ | Quiescent supply current and pin leakage | Bridging defects, gate oxide damage |
| Scan (stuck-at, transition) | Logic structure through DFT scan chains | Manufacturing defects in logic |
| Memory BIST | Embedded SRAM and other arrays | Bit cell failures; also data for repair |
| Memory repair and fuse programming | Maps failing rows or columns to redundant ones | Turns a failing die into a good one |
| Parametric and analog | Regulators, references, PLL lock, ADC/DAC basics, trim | Out-of-spec analog blocks |
| Speed or voltage screening (where possible) | Minimum operating voltage, frequency bins | Slow or weak dice |
Full at-speed testing of high-speed interfaces is often limited at probe because needles and long signal paths cannot carry multi-gigabit signals cleanly. Those tests are usually left for final test or covered by on-chip loopback and BIST.
Why wafer probe matters: a worked example
Wafer probe testing is an important step in the manufacture of chips because it helps to ensure that each device is functional and meets the required specifications. It improves the reliability and performance of the final product and reduces the cost of manufacturing by stopping defective devices before money is spent on them. A simple calculation shows why. The numbers below are illustrative, not from a real product.
- Gross dice per 300 mm wafer: 600
- Probe yield: 85 percent, so 510 good and 90 bad dice
- Package plus final test cost: 2.00 USD per die
Without probe, all 600 dice are packaged, and 90 x 2.00 = 180 USD per wafer is spent packaging and testing parts that will fail. Over a lot of 25 wafers that is 4,500 USD wasted, before counting the lost assembly capacity. If probing costs, say, 60 USD per wafer, it pays for itself three times over in this example. The saving grows with expensive packages: for a large multi-die package, a single bad die can scrap all the good dice packaged with it, which is why chiplet products push for known good die (KGD) testing at probe.
Wafer maps, bins and yield learning
Each die is given a bin code, for example bin 1 for pass and other bins for each failure type. The wafer map is the picture of those bins across the wafer. Engineers read patterns in it: a ring of failures at the edge points to a process uniformity issue, a scratch-like line points to handling damage, and a repeating pattern at the same spot in each reticle field points to a mask or lithography issue. This feedback to the fab is one of the most valuable outputs of probe, beyond the pass and fail decision.
Multi-site testing and test time
Test cost is roughly ATE cost per second times test time. To cut it, probe cards are built to test several dice per touchdown (multi-site). With illustrative numbers: if a single die takes 2 seconds and the card tests 8 dice at once with 90 percent parallel efficiency, the effective time per die drops to about 2 / (8 x 0.9), roughly 0.28 seconds. Memory testing is an extreme case, where very high parallelism per touchdown is common.
Common mistakes and misconceptions
- Treating probe as a full test. Packaging can add defects (wire bond, bump and stress failures), so final test is still needed.
- Confusing probe with WAT/PCM. WAT measures scribe-line structures for the process; probe tests the product die itself.
- Ignoring probe damage. Too much overtravel or too many touchdowns can crack pads or damage the circuitry under them, which hurts later bonding.
- Blaming the die for contact problems. A dirty or worn probe card can show up as a cluster of open-short failures; retest and card cleaning data help separate the two.
FAQ
Is wafer sort the same as wafer probe?
Yes. Wafer sort, wafer probe and circuit probe (CP) are names for the same step.
Can a die be tested at speed on the wafer?
Partly. Core logic can often be tested at speed using on-chip clocks and scan, but high-speed I/O is usually limited by the probe path and is tested more fully after packaging.
What is known good die?
Known good die (KGD) is a die tested well enough at wafer level to be trusted in a multi-die package without a separate packaged test of its own. It needs more thorough probe testing than a single-die product.
Why test hot and cold at probe?
Some defects and marginal circuits only fail at temperature extremes. Testing at temperature on the wafer avoids packaging those dice, which matters most for automotive and industrial parts.
Other articles in this series: https://blog.digitalelectronics.co.in/2023/02/what-are-different-kinds-of-silicon.html



Your comments will be moderated before it appears here.