Semiconductor Manufacturing International Corp. (SMIC) recently announced that the company ceases the production of DRAM at the end of April due to falling DRAM prices. SMIC is reportedly switching production from DRAM to Spansion’s MirrorBit NOR flash memory production and has informed chip-making equipment suppliers and customers of its new direction. SMIC had previously contracted to make DRAM for Elpida and Qimonda, with DRAM production accounted for 23.6% of SMICs revenue the fourth quarter of 2007.
SMIC had previously shifted their NVM effort from Saifun to Spansion late in 2007 and has signed a preliminary memorandum of understanding with Spansion that would allow SMIC to enter selected segments of the flash memory market with a license to manufacture and sell 90nm and 65nm Spansion MirrorBit Quad products into the
SMIC represents a small percentage of the overall memory production however Semico believes that SMICs activities are worth examination for two critical points.
Semico has been unique in insisting that any analysis of either DRAM or NAND supply and demand is incomplete without simultaneously considering the supply and demand for both technologies. There are two tenets that are the basis of that approach.
The first tenet that the production output can be shifted between these two high volume products, and there is ample evidence in support of that belief in the efforts of Samsung and Hynix to shift the output of those two products in order to more efficiently balance the supply and demand for both products.
SMIC’s efforts to shift from DRAM to NOR bears further consideration because it is the first effort that suggests our thesis may extend to all memory production and is not strictly limited to similarities between NAND and DRAM. Some logic manufacturers point to the growing gap between memory speeds and logic speeds while not mentioning that the process objectives of memory companies is to store energy as opposed to getting the energy across the circuitry as quickly as possible. A successful transition by SMIC from DRAM to NOR would emphasis the separation between logic processes that continue to follow Moore’s Law and would point to a higher level of commonality among all current processes whose objective is to contain energy within a storage cell.
The second concept that is critical to our memory analysis is that the basic infrastructure of memory production has undergone a radical transformation over the past decade. One important change is the layers of JVs and partnerships that are the underlying source of production, and SMIC is a perfect example of the options that result. Under this new structure, SMIC’s withdrawal from DRAM production effectively allows both Elpida and Qimonda to reduce their DRAM capacity during this period of oversupply without having the burden of disposing of a fab. Semico sees this new structure of JVs and partnerships as allowing the leading DRAM suppliers more flexibility in adjusting to changing market conditions without causing major ruptures in their own internal organizations and financial structures.
And as for SMIC? Semico’s opinion is that changing SMIC’s production from DRAM to NAND may have a short-term positive impact on SMIC’s revenue due to several considerations, such as fewer distractions than having two different customers on two different cell technologies, and the current oversupply of DRAM products. The longer-term prospects depend on SMIC’s ability to successfully transition to a new technology. We believe SMIC’s previous experience with DRAM will assist them in this transition, and that the relationship with Spansion could give SMIC a competitive edge in future NVM production.