When a company launches a line of memory they have a number of brand choices when it comes to the IC’s that are going to be used. There are also many other manufacturing issues such as PCB design/manufacturing and assembly, but today I am going to cover the IC’s specifically.
Memory manufacturers have a number of variables that they must look at before deciding on which IC they will select to use on their modules. One of the major factors is qualification testing. This is done by getting IC’s from Winbond, Hynix, Samsung, and others for testing on common motherboards to see how they perform
One of the first steps of picking an IC is qualification testing. This is where the memory manufacturer goes out and gets a few batches of IC’s from Winbond, Hynix, Micron, Samsung, and others. They then build modules that will be used for in-house testing on common motherboards and see how they perform. This is also where they look at the characteristics of the IC’s. One characteristic they look at is the CAS Latency. The majority of current DDR500 uses either Samsung or Hynix IC’s. When testing differences between the two IC’s it is clear that Samsung can only run CL3 while Hynix has the ability to run CL3 and CL2.5. If you are aiming a product at overclockers who want tighter timings, it is clear from qualification testing that Hynix IC’s are better for this niche market, but there are other factors to look at before picking the correct IC.
Since IC prices vary from brand to brand, companies must look at the overall performance of an IC found during qualification testing versus the cost of using that particular IC. This is a very critical step because this affects the profit for the company, the markup percentage for the retailer and of course the cost for the end user.
Hynix IC’s have a very special tariff on them thanks to a situation that happened back in 2001-2002! It all began when U.S. memory giant Micron Technology Inc., accused Hynix of benefiting by $11.7 billion in subsidies as part of three bailout deals during the period lasting from January 2001 to June 2002. It has basically been proven by the United States and European Union that due to bailouts from state-owned banks, Hynix has been able to survive in the DRAM market despite massive debts. (Hynix has always denied these claims.)
Starting back in June 2003 the U.S. Commerce Department placed a 44.7 percent tariff on Hynix IC’s when they come into the American market. Shortly after this ruling in August 2003 the European Union (EU) announced a tariff of 34.8 percent on Hynix’s memory chips for the next five years, again accusing the South Korean firm of receiving government subsidies.
Kingston explained to us that since their HyperX line is assembled in America that they have to sometimes pay double tariffs on memory when shipping to Europe. Since Kingston purchases the Hynix IC’s from Asia and ships them into America, and therefore has to pay the 44.7% tariff. Kingston then builds their modules using these IC’s and since they have retailers worldwide they ship some to Europe. Since the built modules contain Hynix IC’s they then have to pay the 34.8% tariff when they arrive in Europe! As you can see this is costly and takes a huge portion of the profit margin away.
Just because an IC is available today doesn’t mean that it will be here tomorrow! Companies release new revisions of IC’s just like any other part made, and this too must be taken into consideration. An example of this would be the Winbond BH5’s (400MHz rated modules that can do CL2-2-2-5 1T) which Winbond stopped manufacturing in September 2003. There were around 30,000 Winbond BH5 IC’s floating around that and OCZ has recently purchased and used to release a "Limited Edition" PC-3500 module based on these IC’s.
Why didn’t Kingston or Corsair pick up these last stockpiles and produce some Winbond BH5 last minute modules? To answer this, think about what happens when a customer returns a bad module using these limited edition IC’s and if the Winbond BH5’s are gone it can not be replaced? Also, it was rumored that there were only 30,000 Winbond BH5 IC’s left in the channel. Simple math shows that 30,000 IC’s divided by 16 IC’s per module equals 1875 memory modules could have been produced. Since the majority of this memory is sold in pairs divide the 1875 modules by 2, which gives you 937 matched pairs of memory.
The above example was done assuming that all 100% of the IC’s make it through the manufacturing and testing process, but trust me there are flaws and the number that make it to the customer will not be 100%. Kingston, almost a 2 Billion dollar a year company, has so many channel partners and customers that producing a new "limited edition" product for only 937 pairs of memory would be silly on their part. The demand would be too high and the possible return issue with discontinued IC’s too great for their reputation. This is only one scenario that companies must look at when picking a particular IC to carry on a specific line.
Now that you understand the very basics on IC selections let’s begin testing the IC’s we are talking about today.