Friday, April 21, 2006
Electrical and Functional Test
Memory within the computer has gradually become the system bottleneck. Microsoft OS demands more and more memory. Fortunately, price of memory has descent. The mainstream memory for this year is 512MB. By estimation, single 1GB memory module will dominate within two years. Although DDR memory is at its old age, it is still running strong. It has not given up any glamour to the new DDR2 memory. While DDR2 has not yet achieved relative market position, DDR3 has appeared. It wants to share the throne. Presently, DDR2 is facing marketing roadblock ahead and also a rising DDR3 from its behind. How long DDR2 can last is a question that nobody has the answer now.
Due to the ever increase in operation frequency, pin count has to be increased also. DDR2 chips are now packaged in BGA package conforming to JEDEC specifications. It looks great with large capacity in a compact package, yet the shortfall is in mechanical durability. Even though DDR2 is a big advance in technology, but its ruggedness is a disappointment. It is marvelous technology, yet for the sake of higher frequency, the onboard resistors and capacitors had become too small and too tight space to handle. Well, on the other hand, since the end users are not likely to insert memory modules into their computer regularly, these problems might not concern them. Moreover, they will soon learn to be gentle to the modules.
Overall, DDR2 does bring many new problems to the module manufacturers. In this fast changing environment, the market will favor those who can solve those problems ahead of others. Obviously, it will require new technology and new methodology to bring new profit.
1. Test environment has to be upgraded
Due to the difference in labor cost and quality standard, there are philosophical difference in the testing of memory module between Asians and the Western world.
In Europe and America, they like to use dedicated memory module test systems. Generally, they are equipped with handler. Automation is highly utilized. Cost of the equipment is high. It can be over a million US dollars for each setup. This figure is un-imaginable to the small and medium size manufacturers. These test equipments not only cost a lot to buy, they also cost a lot to use. It is beyond the manufacturers can bear especially in this extremely low margin industry. Although the more expensive testers can test SDR and DDR, but they run out of steam when testing the higher frequency DDR2 memories. Once the bandwidth is expanded, most machines cannot support due to their hardware architecture. The only alternative is to purchase new dedicated DDR2 memory testers. This cost is the major factor that keeps DDR2 from penetrating the market speedily.
Within Asia, most memory test is still based on motherboard. It is simple and easy. But it is of low tech and low efficiency. It has quality and inter-operative problem. Obviously the advantage is in the reduced cost of equipment. Since cost is the factor that makes Asia the assembly capital of the world, the memory module industry cannot take exception. With the adoption of DDR2 memory, test process solely on motherboard is no longer practical. For the more responsible manufacturers, they might spend about 30 minutes to each to test the memory on the motherboard. (Generally, test time is relative to the test process. Longer test might yield more accurate results, yet it is not certain that over test might not result in over aging also.) Some individual manufacturer would conclude their test just upon “boot up” of the system. This might be acceptable with the previous generation memories. To help to stabilize DDR2 memory, there are DQSN and ODT signals. At this time, no motherboard can unveil the mystery problem on these few signal lines. That’s because that no motherboard were designed for testing memory modules. Module manufacturer usually only utilize off-the-shelf motherboards. With these motherboard tests, even the most honous memory module manufacturer is betting on their luck. Although this is not a well know topic and can be difficult to catch, it would still surface eventually through the propagation of continuous high-speed operation. System will split out a memory error in the middle of your spinning wheel game. Most of these errors are indeed attributed to system memory. Of course, most users can recover by rebooting their system. They can then put all the blame on Microsoft operating system. Frankly, Microsoft had taken so many of these blames so far. Some of the systems might not even boot until the memories are changed. Therefore, I would suggest that you only use name brand memory when you upgrade the main memory in you computer. I would urge you to pay more attention if you have not done so in the past. The problem will become worse as the operation frequency of memory increase. For the memory manufacturers, I would suggest that you do not take short cuts that can destroy all the credit and brand name you have built up. Don’t give your custom a chance to suspect that you module has inter-operative and stability problem.
2. Requires more detail test result
Since DDR2, package has been migrated to BGA, it has been the worst of the worst. Previously, even though frequency is high, we can at least inspect the waveform with a high frequency oscilloscope. But now that the solder balls are hidden underneath the IC, we cannot see nor touch them any more. Once the module exhibits problem we do not know where to start troubleshoot. Of course, some engineers would create test points for measurement during the prototype stage, yet they will only be deleted at the production phase.
Those modules that can boot and be tested on the motherboard normally are rid of internal shorts and opens. Motherboard can usually catch the logical operational faults. Yet, what about the large number of modules that fail to boot in the system? Can we just disregard them or do we change all the chips? These are all valuable money we are throwing away. Well, most of the production tests should be concentrated on finding the cold soldering joints, yet the motherboard test is practically for uncovering bad IC’s. This is a direct contradiction. Memory manufacturers have no good way of detecting the problem caused by manufacturing process. Instead, they can only discover the bad IC. Some manufacturers would use x-ray inspection, yet they can only detect the more serious cases like wrong IC orientation or solder ball collapse. Yet it cannot guaranty the connections are complete. Moreover, all x-rays are manually operated and based on the operator’s visual judgment. Efficiency is rather low.
There is only one answer to test this kind of memory module. Specialized memory tester has to be used to tell the location of the bad chip and the cause of the failure. It can focus to the chip and to the pin in question. It can also tell if it is a solder joint problem or an IC malfunction. With this detail information, the bad module can be quickly recovered at the lowest cost.
Price-performance ratio obviously comes into the play. Is there a relatively low cost and complete test solution for memory modules? How do we evaluate the price-performance? Based on market analysis, the most reasonable price of memory tester for small to medium size industry should be under $10,000 US. Test time for single module should be within 1 minute. Of course, guaranteed test result is foremost important. Longer test time does not always mean more thorough test. Repetitive test is not always the best method. Money is saved with the most optimal test time and algorithm. With DDR2 about to become mainstream, whoever first perfected this technology will control the market dynamic. Let me tell you about a Website: www.simmtester.com. Their content is not only their company product; they have also included much memory information and news. You should find your interest there. Unfortunately, it is only in English for the time being. But I hope they will also be in Chinese soon. From their name, we can see that they had started testing module since the “SIMM” era. They have a long history and a lot of experience and strength. Please see the following picture that highlights their model SP3000 memory tester. Ain’t that cute?
By: Arthur Zhao - Translated by Cecil Ho
Copyright © 2023 CST, Inc. All Rights Reserved