The blurring generations
02/01/1998
The blurring generations
Talkin` `bout my g-g-g-generation - Pete Townshend, The Who`s My Generation
Along time ago - way back in the early 1990s - device generations were clearly separated. For several decades, the industry moved in relative unison through successive generations. Technologists used common device architectures and core processes, so that everyone knew (with a great amount of detail) what was meant by "half-micron." For example, DRAM storage cell designs and interconnect metallization processes could both be discussed in reference to a particular generation.
Decreasing the minimum critical dimensions for a mask set while maintaining the same basic architecture and process flow was termed a "shrink" within a device generation. Thus, with minimal additional process development, the same technology used at 0.5 micron could be used for the 0.4 micron shrink. Since most of the world used the same technology in a given generation, all of the shrinks used the same technology.
The simplicity of clear generational divisions will soon be a thing of the past. As many of the core technologies that have been relied upon for decades cannot be extended, no single, obvious best choice for everyone has emerged. The inevitable tradeoffs between architecture, processing, cost, and reliability mean that different fabs with different product mixes will choose different paths over the next decade. Does the phrase "0.18-micron metallization" mean anything?
The 0.35-micron device generation is probably the last one that can be clearly delineated. From 0.25 micron and below into the deep-submicron realm, the basic process variations within the industry have grown to the extent that it is almost arbitrary to draw the lines between generations.
For much of the 1980s and the early 1990s, DRAM technology drove the rest of the industry and DRAM generations could be used to delineate differences in basic processing. In those simpler times, a 16-Mbit DRAM was made with 0.5-micron technology (or a shrink thereof). Today, a 64-Mbit DRAM could be built with 0.35-micron technology, or perhaps with a 0.25-micron process. Also, companies have announced that 128-Mbit chips will be produced; the same basic process flow could produce 64-, 128-, or 256-Mbit die. DRAM generations are no longer a common measure of technology.
The new 1997 National Technology Roadmap for Semiconductors (1997 NTRS) increases the blurring by the definition of a possible 0.15-micron generation. Apparently required by expected delays in the development of a complete 193-nm lithography process (and by a need to relax after the breakneck pursuit of 0.18), 0.15 will be a relatively easy first introduction of that wavelength before extensive mask wavefront engineering will be needed for 0.13 micron.
The result of this blurring is that it will become harder for us to talk to one another about technology. We will all have to become accustomed to allowing more time to explain what structures or processes we are concerned with. Discussing unit-processes may not be too difficult, but talking about process integration (composed of many different individual unit-processes) may be tough. We may have to get used to listing all of the specific unit-processes in describing an overall sub-flow.
For fabs, the blurring of generations means that it may be more difficult to choose new materials and equipment. For equipment and materials suppliers, the blurring of generations will probably make it more difficult to assess overall customer requirements in allocating development and applications resources.
Due to the conceptual momentum of several decades, it is certain that we will still talk about "generations" for many years to come. The 1997 NTRS is extraordinarily valuable as a guidebook to future technology requirements, and the Roadmap is still laid out in generations. Even as future technology directions are uncertain, the most common term used in the industry - the generation - will increasingly represent uncertainty.