by Scotten W. Jones, IC Knowledge LLC
Editor’s note: This letter is in response to "The economic realities of 450mm" by SEMI’s John Ellis.
February 19, 2010 – The possible transition to 450mm wafers would be an enormous undertaking for the entire semiconductor industry ecosystem. Developing the best possible understanding of the trade-offs is critical to good decision-making. A free and open debate about the economics of the transition benefits everyone. I enjoyed reading John Ellis’ article and also following the included links to read some of the material on which it is based. However, I would like to respond to some of the statements made therein, as well as point out some mischaracterizations of my work.
Mr. Ellis points out that I have assumed that 450mm tools will have throughput identical to 300mm tools. He goes on to state that "beam tools" will not be able to meet this assumption based on "physics," and cites SEMI’s Equipment Productivity Work Group’s (EPWG) paper on the economics of 450mm for the basis of his statement. I had not previously seen this paper, although I have now read it. Unfortunately the EPWG report only states the throughput without providing any detail that would allow outside parties to evaluate the validity of the conclusion. I do agree that throughput is a critical assumption in my analysis — and later in this letter I will present some calculations using the lower throughputs cited by the EPWG report to show the impact that lower beam tool throughputs would have on cost.
Mr. Ellis states that "IC Knowledge’s assumptions on consumables cost and wafer costs are also not supported by analysis," and "all of the optimistic predictions of cost savings accrued by going to 450mm wafers assume that 450mm wafers will cost approximately the same per unit area as do 300mm wafers." Both of these statements are wrong and do not correctly characterize the analysis that was done for the "300mm Prime and the prospect for 450mm wafers" article. Historically each new wafer size has a much higher cost per unit area for the starting wafer and then over time the cost per unit area comes down until it is close to or equal to the more mature wafer sizes. I have assumed that the starting wafer cost per unit area for 450mm wafers is approximately 5× the cost per unit area of 300mm wafers in 2015, and that assumption is reflected in my analysis presented in the paper. Furthermore, I have assumed that water, energy, gases, chemicals, and other consumables also show increased usage with the larger wafer size and those assumptions are also reflected in my calculations.
Mr. Ellis goes on to say that "IC Knowledge has stated that 300mm Prime will not meet its productivity improvement objectives," this also is not an accurate characterization of my article. What I say in my article in two places is that to-date 300mm prime has not met its productivity goals. I also point out that no one has been able to outline a path to do so, and that this leaves 450mm "on the table."
I would also like to point out that in the writings I have done about 450mm, that while I have concluded that it will produce a wafer cost saving for semiconductor manufacturers, I have also pointed out that the economics for the equipment companies do not look favorable.
The work I have done on 450mm is based on simulating costs using my commercially available IC Knowledge — IC Cost Model. One of the strengths of the model is the ability to easily simulate different assumptions. I have a version of the model modified for 450mm calculations where for each type of equipment – cost, throughput, and footprint multipliers versus 300mm may be set as well as energy, water, chemicals, gases, and other consumables usage. The current ISMI guidelines for 450mm call for no increase in consumables per 450mm wafer vs. 300mm wafers, and also for tool footprints to be the same for the same throughput (taller is OK within reason, or larger is OK if throughput is correspondingly higher). To the best of my knowledge, ISMI has not set tool cost targets. I have used the 450mm version of the cost model to simulate three sets of conditions for a 22nm process produced in 2015. The first is 300mm; the second is 450mm with 1.0× 300mm tool throughputs, footprints and consumable usage; and the third is 450mm with the beam tool throughputs and footprints adjusted based on the EPWG numbers (the footprints are adjusted down in proportion to the throughput reduction). For 450mm I have assumed an average 1.3× increase in tool cost with exposure and some other tools >1.3× and other tools <1.3×. The resulting improvements in wafer cost per unit area are 29% for the 1.0× assumption and 14% for the EPWG assumptions. Please note that in both cases these are 2015 costs that include an approximately 5× cost per unit area increase in starting wafer costs. Over time the starting wafer multiplier should come down further, increasing the cost savings to as much as 37% in the first case and 22% in the second case. As additional projections around tool performance, material cost and other wafer fabrication parameters are developed the 450mm cost modeling tool provides a powerful technique for evaluating the impact on wafer cost.
The 450mm transition continues to be a hotly debated topic. Based on my work to date, I continue to believe that 450mm will produce a wafer cost reduction for semiconductor manufacturers. If 450mm tool throughputs of 1.0× are achieved, then a larger cost reduction results; if lower throughputs are achieved then the cost reduction is smaller. Clearly, companies such as Intel and Samsung agree with this conclusion, or they wouldn’t be pushing so hard for it. From an equipment company perspective, I believe 450mm is problematic with return on investment uncertain at best consistent with the reluctance of most equipment companies to support 450mm.
Scotten W. Jones received his Bachelors of Science from the U. of Rhode Island, and is a senior member of the IEEE, a member of the Electrochemical Society, and is president of IC Knowledge LLC, P.O. Box 20, Georgetown, MA 01833 USA; 978-352-7610; firstname.lastname@example.org, www.icknowledge.com.