ATE Choice Demands Careful Evaluation

Choosing the best system requires more than buying the latest model or looking for the most features per dollar. It requires knowledge of tester technology and test-industry business conditions.

Carl Duffy and Michael Raico, Advanced Micro Devices, Dresden, Germany -- Test & Measurement World, 5/1/2000

With typical prices of $500,000 to $3 million each, semiconductor ATE systems are probably the most significant capital investments a semiconductor company will make outside of its fab equipment. The choice of ATE is also one of the most significant financial decisions you are likely to influence. Evaluating the features of a tester is difficult for managers who do not personally use testers day-to-day, so the choice of tester is weighted heavily on test engineers’ opinions.

The importance of the decision stretches beyond purchase price. Most testers are not isolated purchases. There are other considerations, such as accessories and software. When you need additional testers, it is usually wiser to buy duplicates of the model you already have rather than look for different models. As you add duplicate testers, they will accommodate your stock of load boards and probe cards. In addition, your programmers, maintenance personnel, system administrators, and operators won’t need retraining. Also keep in mind that identical production machinery can reduce bottlenecks because you can redistribute workloads to other machines when one machine is down.

Comparing Road Maps
Settling on a tester family is a three- to five-year commitment—it is such a large commitment of time and money that a poor choice will result in problems visible to all. When evaluating a new test platform, consider two road maps: your company’s road map and your tester vendor’s road map. Understanding your product line’s road map (Fig. 1) is essential. You want the right solution for your current product line, and you want to satisfy requirements for future products, based on inputs from your organization’s sales, marketing, production, and engineering staffs.

Your product-line roadmap should plot specific technological and packaging changes you expect to see in your products as they evolve throughout the expected lifetime of the tester you ultimately decide to buy. The map should show clearly the dates at which you anticipate significant changes in these key areas:

• package style (for instance, gull-wing-lead to multi-chip module),
• package pin counts,
• device operating frequency,
• power-supply voltages (for instance, lower or multiple voltage levels),
• noise performance,
• process geometry, and
• design-for-test techniques (for example, the adoption of full-scan or built-in self-test techniques).

Your road map should also indicate whether you’ll need to test special devices like a Rambus chip, a chip that contains multiple clock domains, or an SOC. A leading driver of SOC technology is core reuse to improve time to market. But if you need weeks to adapt your tester to SOC test, then you’ve lost the rationale for SOC. If you don’t know enough about technologies like SOC and Rambus, ask vendors whether their systems can test these features.

Figure 1. Your product road map describes how your technology will change to meet the demands of marketing within the constraints of engineering and production know-how. Work with a tester vendor whose own road map parallels yours for the three- to five-year life of the tester you’ll purchase.

Technical Issues
When evaluating testers, explain your concerns to the vendors. Address your critical performance questions to applications engineers or to the tester’s designers. You need first-hand information that salespeople can’t always provide. For example, is analog instrumentation available that can directly make the tests required for your mixed-signal devices, or will the test require extra circuitry on the load board to compensate for instrument limitations?

Consider also the data-processing speed of the tester. If a half-price tester takes twice as long to test something, then you will gain nothing by buying it. Consider also a tester’s ability to test multiple parts in parallel (see “ Are Two Heads Better Than One?” see right).

Getting Advice
If you know test engineers or purchasing agents at other companies, contact them and get their opinions. Go to trade shows and meet with former colleagues. Ask about their experience with service and the applications support.

To understand the advice you’re getting, learn the buzzwords, like “vector memory depth” or the aforementioned “tester per pin.” Realize that each vendor has its own interpretation for each buzzword—be sure to ask for its interpretation.

 For example, vector memory is the part of the tester that stores the digital data fed into a UUT to exercise it. A tester with insufficient vector memory may have to reload vector data off a disk for every part it tests, needlessly lengthening your test time. The best way to estimate memory requirements is to look back at old test programs and see how much memory you’ve used in the past. Draw a graph of vector memory used (vector depth multiplied by number of digital I/O pins) versus gate count and extrapolate to estimate what you’ll need for future designs (Fig. 2 ).

Figure 2. You can extrapolate from earlier projects to estimate your future test-vector requirements. Be sure to account for changes in pin count.

Beware that a vendor’s interpretation of vector memory might not match yours. The vendor may tell you that its vector or scan memory is X megabytes deep, but you may not have full use of that memory because of restrictions that depend on the tester architecture and your device’s architecture. For example, one company’s tester requires your UUT to have 2ⁿ scan chains to make full use of all the scan-vector memory. If you have an old part with three chains rather than two or four, you can only access 75% of the tester’s total scan memory. Having scan chains of different lengths can also waste some of the scan memory.

While at trade shows,¹ make your presence known to ATE vendors. They will offer quick technical demonstrations of their products, which can help you narrow the field to a few testers. Then, pay a visit to the finalists’ factories and meet with their in-house technical experts.

At the factory, let the vendor drive the tester for you. Again, you will see canned demonstrations, but in a one-on-one environment you can see more of the nuts and bolts of the tester and test development. Ask the demonstrator to deviate from the presentation, making modifications on the fly. On a digital tester, ask to see how to modify pin timing. On a mixed-signal tester, have the demonstrator create a new waveform. Have her or him analyze data in the time domain and in the frequency domain. Ask to see instrument-to-instrument tests—for example, connect a waveform generator to a waveform digitizer. Consider how efficient an engineer would be using these tools.

If you will be creating digital tests from either simulation tools or ATPG (automatic test-pattern generator) tools, understand how the data will get from tool to tester. Unfortunately, the vendor may have less experience generating patterns than you do, and the vendor may not have exposure to the software your DFT team uses to create test vectors. Ask if you will have to create your own data-conversion programs or if the vendor or a third party offers a data-conversion tool to create test vectors from your ATPG and simulation-tool data.

You need to understand how this new test platform fits into the current production environment. What is the operator interface like? Is the operator interface compatible with the current methods being used in your production sites? Will operators need training on these testers? Are these testers compatible with your handlers and wafer probers?

Most vendors provide the necessary software and hardware to allow various handlers and probers to interface with their testers both electrically and mechanically. Will the test head mate to your handler or prober without serious modification to the equipment? Make sure that there are test-head manipulators available that will handle this test head. Can you get the test data into your factory databases, or will that require data-conversion software development? Does the tester support STDF (Standard Test Data Format), an industry standard data format invented by Teradyne?

You’ll face additional costs and delays if you or your vendor, or both of you have to develop any of the software that interfaces tester data to other factory applications. The vendor may be interested in developing an interface to your handler if the vendor could then sell the interface to other customers. Intriguing as this offer may be, you should be aware that your test project will be on hold during the vendor’s development effort, which could in turn face unforeseen delays.

The Deal
When it comes time to strike the deal, understand the prevailing business climate in the ATE industry. Most large corporations have procurement specialists, some of whom may deal only in ATE procurement. If your company employs such people, let them get involved. They will understand the current state of the ATE business. Is the vendor hungry for your business? Will this vendor be in business next year to service your machine or sell you more machines?

Make sure the procurement people understand your needs and requirements. This means getting them involved early in the process. They can help you negotiate extras into the deal. Also, make sure the contract specifies the price for future purchases. If you are intimidated by the concept of negotiating, or if you think negotiating is a case of two parties naming a price and then meeting somewhere in the middle, read a book on negotiation.²

If you represent a small company with limited financial resources, you might consider purchasing a tester for test-program development while using an outside test house for production test. Find out what test platforms the test houses you’re considering commonly use. Most tester vendors will share their installed customer database if you show an interest in using a test house for your production site.

If your technology is a new line of business for your ATE vendor, then perhaps it will want to develop in-house expertise so it can support this technology for other customers. In that case, push for your vendor to develop your first production test program. Agree on an acceptable range for the production test time per chip, and make that spec part of the deal. With a vendor-generated program, you run less risk of delaying release of a new product or of shipping bad parts to your customers because of your inexperience with a new system. The vendor’s involvement makes it a partner with a strong interest in ensuring the tester is technically capable of meeting your current needs.

Each vendor will display its product in its most favorable light. Make sure when reviewing data that all things are equal. Most benefits come with tradeoffs; make sure you understand everything the tester can and cannot do. Follow the guidelines we’ve provided, and you’ll find you’ve bought a tester that continues to satisfy your needs as your products evolve. T&MW

FOOTNOTES
1. “Show & Meeting Calendar, ” Test & Measurement World.

2. Fisher, Roger, et al., Getting to Yes: Negotiating Agreement Without Giving In, Penguin USA, New York, NY, 1991.

Carl Duffy is a senior product and test engineer for Advanced Micro Devices’ Communication Products Division, Austin, TX. He is currently on assignment in Dresden, Germany. E-mail: carl.duffy@amd.com.

Michael Raico is a product and test engineering manager for Advanced Micro Devices’ Communication Products Division, Austin, TX. Raico is on assignment in Dresden, Germany. E-mail: michael.raico@amd.com
 

Will This Product Perform After You Buy It? Consumer Reports has yet to bring out an annual “ATE Buyer’s Guide” edition, so you will have to evaluate potential quality other ways. Two common metrics that vendors will quote are field MTBF (mean time between failure) and MTTR (mean time to repair). Your own production staff may have this data for the testers you already own.   Bear in mind that how close you are to service personnel may influence which of these two metrics will be your primary criteria (see the figure). If you have service personnel available 24 hours per day, you may be more influenced by MTBF because it will reflect your downtime. If the nearest service personnel or replacement parts are in another country, you may be more interested in getting guarantees of low MTTR from the vendor. Vendors will quote either MTBF or MTTR figures. You need to decide whether long up times (yellow) or short down times (red) are more important to you.   Either way, you can try to negotiate guarantees into the purchase contract. Ask what concessions the vendor can give if MTBF or MTTR doesn’t meet its specification; perhaps the vendor would extend the warranty or give a discount on the maintenance contract. You should also discuss up front the “lemon criteria”—how you will decide if a tester is not meeting expectations and needs to be replaced. Vendors will be very reluctant to replace a tester because of both cost and public reputation.   Consider also the quality and cost of the maintenance, not just during the warranty period, but also for the maintenance contract you will purchase after the warranty expires. Where are the nearest service personnel? How much does 24-hour coverage cost you? Do you need 24-hour coverage? To guarantee faster repair, the vendor will keep some spare boards in stock locally; that’s the real reason why you pay more for fast repair. The cost of this will obviously go down for the vendor if there are many of these testers in your locale.   Question the vendor about who does the maintenance. How many service personnel are in your region? Is maintenance subcontracted? If so, what kind of relationship exists between vendor and subcontractor? Does the subcontractor have to keep MTTR down to get paid or is the subcontractor paid based on the hours he or she spends on your premises?   Finally, ask the vendor for a tour of its production facilities. See what quality-control measures are in place. Is there a visible feedback system that focuses on continuous improvement? Ask some of the production-line employees questions to find out how they are participating in continuous improvement. Satisfy yourself that the charts on the wall aren’t just a public relations exercise.—Carl Duffy and Michael Raico