Forecast: Memory Faces Hard Times, 450 mm Moves Further Away
Alexander E. Braun, Senior Editor -- Semiconductor International, 12/7/2007
Unless it gets an infusion of $65B from outside sources to fund capex expenditures through 2011, the memory industry will face great difficulties. However, with the possible exception of government grants, everyone else will demand more financial restraint and profitability than memory device sales can produce. This was an assessment made by Bob Johnson, research vice president at Gartner Dataquest (Stamford, Conn.), during his presentation, "Financial Challenges Facing the Semiconductor Industry," at Gartner’s Semiconductor Industry Update, held this week in San Jose.
“Before 1995, the industry could generate both increasing ASPs and increasing volume,” Johnson said. “Since then, more traditional economic ASP/volume relationships have emerged, with lower ASPs accompanying increasing volume.” From the mid-1960s until the mid-1990s, the semiconductor market grew at a long-term rate of between 15% and 20%. However, a DRAM-induced market crash, the Asian financial crisis in the late 1990s, compounded by the dot-com annihilation in 2001, slowed long-term growth to between 10% and 15%. Several factors have combined to keep downward pressure on the industry’s long-term growth prospects. The lack of a "killer application" to rival the PC or cell phone forces the industry to rely for growth on average selling price (ASP)-sensitive digital consumer electronics applications.
“Over the past decade, the decline in prices and industry revenue per square inch of silicon have put pressure on industry costs and profitability,” said Johnson, adding that the revenue decline per square inch since 1995 is roughly 20% and that, so far, manufacturing costs have handled this drop. “The move to 300 mm wafers generated a ~30% cost savings for advanced manufacturing, while fab productivity improvements helped maintain industry margins,” he said. “However, with manufacturing costs for new processes increasing, pressure on margins won’t abate.”
R&D expense for process development, fab equipment prices and design costs are rising faster than the industry’s revenue growth. Maintaining R&D at a fixed percentage of total revenue per established business models will not generate enough funds for process development and new device design at the rate of a new node every two years unless costs are reduced or shared.
Capital investments by memory markets have dominated the industry investment environment. Partially fueled by the NAND flash market and soaring DRAM demand, the market requires continued capital investment in excess of 40% of revenue to maintain currently projected unit growth. However, at projected unit growth rates, ASP declines guarantee that total revenue will remain static; huge investments become necessary just to break even.
Traditionally, semiconductor manufacturers have responded by adopting strategies to keep costs in line with revenues and maintain profitability. Among these are focusing R&D efforts on company strengths and outsourcing everything else, investigating design alternatives (i.e., packaging) instead of automatically going to the next design node, slowing down the adoption of new technology, partnering to share development costs, and joining R&D consortia.
A basic financial model for the industry was presented, with the emphasis on cost of goods sold (COGS), depreciation (also included in COGS), R&D expense and net income. It was based on an analysis of publicly held semiconductor companies that account for roughly 60% of total industry revenue. The analysis period was from 2003 to 2006, which represented a time of reasonable industry performance and profitability, and covered the basic structure for the entire industry and selected major device segments. The logic IDMs were the most profitable segment. However, this includes Intel (Santa Clara, Calif.), which alone accounts for almost one half of the profits.
The model is a means to estimate whether the industry can generate sufficient cash flow to fund its needed investments. Future projections were based on an historical model applied to forecasted revenue and capex. The model showed that while the industry as a whole can generate the funds for needed investments, it is not certain that every segment can do so.
“In memory’s case, the fact that capex has increased dramatically over the years must be considered,” Johnson said. “Since projected revenue is relatively flat, depreciation as a percent of this revenue will increase dramatically. By 2011, there will have been five years of virtually flat revenue combined with five years of extremely high capex. By applying this model to net cash flow and modifying the basic memory financial model to account for increasing depreciation costs, it appears that the memory market’s cash flow will be very problematic.”
R&D faces similar problems. The model projects an annual growth rate of R&D spending of ~6% per year. This is insufficient to fund new technology at historical rates or the R&D to develop 450 mm toolsets over the 2008-2011 time frame. Some semiconductor manufacturers claim that the move to 450 mm wafers is necessary to continue cost reductions and follow Moore’s Law. While there are differences in opinion about the industry’s capability to realize the projected cost savings from 450 mm, the concern over who will pay for the equipment R&D needed to make this transition a reality is fact, not opinion. The device industry, led by Intel and Sematech (Austin, Texas), is pressuring OEMs to fund 450 mm R&D. The equipment industry is pushing back, protesting that it never got an adequate return on its 300 mm equipment investment.
The Figure shows OEMs’ capability to generate funds for 450 mm development over the 2008-2011 period, necessary to meet the 2012 450 mm production target. The projections also include R&D to maintain the current pace of technology development for 300 mm equipment. It also shows that the projected total net income would be slightly larger than the additional R&D expense needed for 450 mm. In short, without assistance from the rest of the semiconductor industry, OEMs would have to forgo profits for four to five years to fund 450 mm development.
| The semiconductor equipment industry cannot afford the R&D necessary for 450 mm. (Source: Gartner Dataquest) |
As one of the analysts commented, “Investors would hold a necktie party for any executive contemplating this.”
