The term handler seems innocuous enough. It’s not until you consider the total range of technically difficult and often incompatible requirements expected of these industrial robots that you realize how remarkable they are.
In a semiconductor manufacturing process, the so-called front-end activities produce diffused wafers. Each device on the wafer is tested, the wafer is singulated, and the good devices are packaged. The so-called back-end activities include burn-in and test of the packaged devices, followed by grading and marking.
But once the wafer is singulated, thousands of separate devices have been created, and each one must be handled individually. How is a device presented to the tester? How are devices tested at different temperatures? How are the many types of packages accommodated? These are just some of the situations that handlers address.
Main Attributes
Because the handler must physically manipulate the devices, package type, molding and lead tolerances, and reference surfaces are important variables. On the other hand, the tester only is concerned with device functionality. It’s no wonder that a tester capable of driving and measuring the hundreds of very high-speed signals associated with a complex logic device is the highest-cost back-end piece of equipment.
Reliability
The handler feeds devices to the tester. As a result, reliability has become one of the basic requirements for a handler. You don’t want to have line stoppages caused by the ancillary equipment that only supports and augments the tester.
ICs aren’t the only type of device to be tested automatically and positioned by handlers. While discussing the development of a new memory module handler, Cecil Ho, president of CST, commented that users of existing handlers were unhappy with the present contactors. “Our engineers looked at plugging modules directly into a test socket. It not only solved the contactor problem, but it also removed the ambiguity between manual and automatic test because the same socket was being used for both. However, the engineers knew they had to anticipate the question ‘How often do I have to change the socket?’.”
Through rigorous testing, the engineers established that a conventional test socket would provide up to 200,000 insertions and withdrawals—well beyond the socket manufacturer’s guaranteed 10,000 manual insertion cycles. The large difference was due to the controlled forces applied when the socket was used in an automatic machine application compared to the greater variation of force and speed of manual operation.
“With the result [of testing] on record, the engineers concluded that the socket would work for more than 100 days in the normal production environment,” he explained. “Together with the socket receptacle to allow quick replacement, it would be a much more cost-effective method when compared to the conventional contactor.”
Speed
Even if your handler is reliable, it still can present a bottleneck. The index time must be short enough that it causes minimum delay beyond the time the tester requires. John Pollack, vice president of corporate product marketing at Aetrium, said, “The name of the game is tester utilization. The testers have become the largest cost item for the back end, so manufacturers do whatever they can to keep them up and running.
“Users have become very creative and that has caused the handler companies to look at and define both industry and customer needs,” he continued. “What the manufacturers are after is more throughput. For example, you need to figure out how to give them shorter setup and changeover times.”
When trying to determine handler throughput, according to Dennis Nelson, executive vice president of sales and marketing at MCT, “A customer should ask how long it will take to load, test, and bin 10,000 parts. For this exercise, binning can have a specific bin split and then must be followed by the time taken to put the parts back into tubes or trays or attach them to tape.”
Another way to improve test times is to test multiple devices in parallel. Several manufacturers support up to four or eight sites, and a few offer as many as 32 or 64.
Mr. Nelson added, “A new manufacturing strategy is developing as a result of a new, nonsingulated handling concept. Panel/strip handling is a significant means to dramatically change the cost of test. Throughput improves by two to four times over that of gravity handlers and by up to 10 times over pick-and-place machines.”
As part of the assembly process, good die from tested wafers are mounted on lead frames which are part of a standard-size panel or strip. Each panel/strip contains a few to a few hundred devices. The key to higher throughput is to test them before the panels/strips are singulated.1
Temperature Testing
Unfortunately, all ICs are temperature sensitive, and to meet guaranteed device specifications, testing is performed at a number of different temperatures. Presenting the tester with hot-, cold-, or ambient-temperature ICs is another of the handler’s jobs.
Conventionally, trays or tubes of devices were soaked in a separate chamber at a controlled temperature. Then they were inserted into the contactor or socket, tested, and returned to a second or third chamber for subsequent testing at another temperature. This practice not only multiplies the length of the test, but also complicates the mechanics of the tester and increases the chance of device damage because of more handling.
The situation has been exacerbated recently by large, fast devices that exhibit significant self-heating during at-speed functional test. Also contributing to the problem are the low thermal-mass interconnection schemes, broadly referred to as chip-scale packaging. Figure 1 demonstrates the lower yield that can result from these factors.
In this example, 2,500 parts have been tested. The blue curve shows that 317 devices can be classified as 480-MHz parts when the die temperature is held constant within 2°C. When the die temperature is less well-controlled and increases 20°C above the desired set point because of self-heating, the distribution curve (the red curve) shifts toward lower maximum operating speeds. The effect reduces the number of highest speed 480-MHz parts to 77.
Because of their low thermal mass and high power dissipation, some logic devices also must be dynamically temperature controlled to avoid lost yield. Even if the package temperature is held constant, the die temperature will vary greatly from the set point. Consequently, several handler companies have developed control systems that attempt to hold the device die temperature constant during functional test.
Schlumberger’s approach is called Power Balancing™. Figure 2 contrasts the die temperature that results from forcing the temperature of the package to be constant and actually compensating for the instantaneous DUT power dissipation.
In actual use, the IC will be subject to self-heating and will not be part of a temperature control system that holds its die temperature constant. In that sense, allowing self-heating to increase the die temperature is more representative of the actual operating conditions of the IC.
However, it’s important to know with certainty how the IC performs at temperature, and power balancing eliminates the variable of self-heating from the test equation. As a result, the manufacturer can guarantee the maximum junction temperature at which the device will operate.
But, beware. Power balancing reduces an IC’s performance margin beyond the guaranteed specification. It is worth considering the implications of a test method that increases yield by almost 300% in the given example.
The 77 devices that could be graded as 480-MHz parts even though the die temperature increased by 20°C actually were the best of the 317. When die temperature is allowed to rise, the test becomes more stringent. Yield reduces because only those parts with a significant performance margin beyond the guaranteed level will perform correctly. In Figure 1, the 77 devices are the furthest to the right near the end of the blue curve.
To use one of the 317 chips, you must derate the specified performance by accounting for the die-to-package and package-to-heatsink thermal resistances in the usual way. You don’t know if you have one of the 77 very good or one of the other 240 devices, so you can’t tell how much operating margin a particular device has.
Taking into account die self-heating in calculations is nothing new to circuit designers. But because of the way the part has been tested, greater than 75% chance exists (240 out of 317) that little operating margin remains beyond the guaranteed junction temperature. To put it another way, if only the case temperature is held constant, almost three-quarters of the devices the manufacturer classifies as 480 MHz parts would fail the high-temperature test.
You are getting what you have paid for but not receiving as much extra as you were in the past. Make sure you completely understand the guaranteed specifications of the device before committing to a design. And don’t assume any performance margin that isn’t actually found on the data sheet.
Besides enabling constant junction-temperature testing, active heat exchangers also provide very short soak times. “Active conduction heat exchangers inherently possess many thermal control advantages over conventional approaches,” explained Joe Hovendon, a product manager at Schlumberger. “Forcing a device to temperature via direct conduction is extremely fast compared to convection soaking.
“The soak chamber size can be reduced because there no longer is a requirement to soak large quantities of devices simultaneously,” he continued. “The reduction in chamber size means a reduction in work in progress (WIP) which, in turn, simplifies the design complexity of the system automation architecture. Active conduction allows multiple set-point testing to be performed while the device is in the test contactor and eliminates the need for repeated cycling of the devices through the handler.”
Some of Aetrium’s recent handler products have similar capabilities but operate in a subtly different way. Jim Roblee, director of engineering for the corporate engineering division, said that temperature is sensed very accurately directly at the handler/device interface. Die self-heating can be compensated for because the power flowing into or out of the thermal head can be measured without an electrical connection to the power supply of the device. However, should a tester manufacturer make those signals available to Aetrium, then power can be calculated directly and appropriate corrections made much as Schlumberger does.
“We’re recording a ramp gradient of about 27°/s that allows us to rapidly change the device temperature from -55°C to +155°C,” Mr. Roblee continued. “This performance can be individually controlled at each of up to 32 sites. We currently have a head with a ×8 configuration running µBGA devices, but the system has been designed for up to ×32.”
Additional Functionality
In an attempt to reduce back-end floor space, some handler manufacturers integrate features such as laser marking and in-line binning into their machines. Opinions differ on the wisdom of this approach.
Aetrium’s Mr. Pollack said, “Any of the competitors can make a product that only handles the device once. The device is tested, inspected, marked, and binned all in one station. However, if the inspection camera or the marking capability suddenly goes down, the tester sits idle until the handler maintenance is completed.”
Schlumberger’s Mr. Hovendon elaborated. “Expanding the automation level of test handlers to include marking and in-line binning should be done in test cells where it doesn’t place an artificial cap on system throughput. As device test times continue to decrease, the handling system has less time to perform functions that are not directly tester-support-related. As a basic guideline, the process scope should not be increased at the expense of throughput or reliability.”
Binning may be restricted by the number of different grades that can be supported by a particular handler, according to Ed Helm, handler product manager for Advantest America. “In-line handling and grading simplify the production flow but also restrict the granularity of tested-product results. Handling equipment with more complex sorting and physical considerations for multiple binning takes extra floor space and slows production throughput. Full software tray mapping allows the best combination of approaches to be used.”
Taking a very positive stance was Mr. Nelson of MCT: “Integrated operations and handling are becoming more important as the back-end [activities] look to become more cost-effective and responsive. Combining process steps such as test, marking, and inspection will eliminate WIP and help to make the overall operation more self-contained, manageable, and profitable.”
Reference
- Nelson, D., “Chip Scale Devices: Handling the Small Stuff,” Micro Component Technology, paper presented at Semicon West 1999.
Published by EE-Evaluation Engineering
All contents © 2000 Nelson Publishing Inc.
No reprint, distribution, or reuse in any medium is permitted
without the express written consent of the publisher.
May 2000