According to a study by Google, there is a huge problem for DRAM. ''Errors in dynamic random access memory (DRAM) are a common form of hardware failure in modern compute clusters. Failures are costly both in terms of hardware replacement costs and service disruption,'' according to the study.

''For example, we observe DRAM error rates that are orders of magnitude higher than previously reported, with 25,000 to 70,000 errors per billion device hours per Mbit and more than 8 percent of DIMMs affected by errors per year,'' according to the study.

The study was done with ''commodity servers'' over 2.5 years. The collected data covers multiple vendors and DRAMs. That study can be read here.

It is unclear which vendors are at fault. One vendor, Lexar Media Inc., has proved to be more reliable than others, said G. Dan Hutcheson, president of VLSI Research Inc. Lexar is a wholly owned subsidiary of Micron Technology Inc.

Hutcheson also said that there is also an inordinate amount of failures for NAND devices, causing problems for USB flash drives and other products.

The reason for the memory failures is that OEMs, test houses and chip makers have cut back--or taken shortcuts--on the memory test process in an effort to cut costs, he said.

The problem is that ''no one wants to pay for test,'' but the practice has backfired in the consumer market, he said at the ISMI Symposium here. This is causing what the analyst calls ''techno-stress'' or a condition that end-user products are difficult to learn or will fail.

And the problem may get worse amid a sudden upturn in the IC industry. Leading-edge fab and overall IC-assembly utilization are both at the 96 percent level right now, he said.

VLSI Research has not changed its overall IC forecast, which is minus 13.4 percent for 2009. What has changed is the demand picture, which is sizzling after a huge downturn in the first quarter.

''The market is hot,'' he said. ''We should see a good Christmas season.''