San Francisco After months of intense and reportedly often heated debate, the Compact Model Council has selected a new industry standard CMOS transistor model. Jointly developed by Pennsylvania State University and Royal Philips Electronics NV, the model succeeds the BSIM3 and BSIM4, which have been the industry standard for years.
The Penn State Philips (PSP) model is a surface-potential model. Surface potential is regarded as a better approximation of the physical behavior of transistors and yields better predictions of the performance of ICs than alternative models. It allows the model to take into account gate leakage and quantum-mechanical effects, which become increasingly important with the downscaling of CMOS technology. Backers say the model is based on the underlying physics of CMOS transistor operation, enabling it to operate with fewer parameters than other models.
Though technologists consider BSIM4 to be a very good model at 90 nanometers and below, weaknesses in BSIM4 can cause problems in some analog applications, even at the 180- and 130-nm nodes, Josef Watts, chairman of the Compact Model Council (CMC), said. These problems arise, Watts said, because BSIM3 and BSIM4 are based on the threshold-voltage concept.
As a surface-potential model, PSP does not use certain assumptions that a voltage-based model does. Thus, the PSP model is a more physically correct model, especially for operation around zero bias on the source and drain or with the gate bias near the threshold voltage, Watts said.
The CMC set up a shootout last May between the PSP model and another model, the HiSIM-RF surface-potential Spice model developed at Hiroshima University, by selecting them as the two finalists for standardization. Both camps fought hard for their respective models, and the PSP model was ultimately selected by a vote of 17 to 14 among the 31 members of the CMC.
People familiar with the council dealings say the debate was often very contentious, something that Watts said is pretty common in cases where the council is choosing between two or more options. "By the time a model is mature enough for us to make it a standard, somebody has invested a lot of energy in it already," Watts said. People fight hard for their models because they have a vested interest in seeing them succeed, he said.
Several issues surrounded this particular debate, Watts said, including the fact that companies based in Japan or the United States largely sided with the model that was developed in their respective countries. Ultimately, he said, the debate came down to a philosophical question the HiSIM model is based on an interactive solution while the PSP model is based on an analytic solution. "Both models produce a very good result," Watts said, "but there are philosophical reasons for doing one or the other."
Reinout Woltjer, head of Philips Research's device modeling group, said in a statement: "As CMOS takes on new roles beyond the production of purely digital chips, it is important that the industry has a single model that accurately predicts transistor performance under all circuit conditions, including RF and analog circuit behavior. By basing the PSP model on the fundamental physics of transistor operation, it provides extremely accurate results over the entire operating spectrum from dc to well in excess of 50 GHz."
After choosing the PSP model, the council next debated whether to distribute the model in C, which is how the BSIM models have been distributed for years, or Verilog-A. The council ultimately chose the latter, which did not please everyone.
Kenneth Brock, vice president of marketing at Silvaco, which sponsored and pushed hard for the HiSIM model, described distributing the model in Verilog-A as "going into some uncharted waters."
But Watts and others downplayed the impact, noting that the PSP model will be available in C through Philips' Simkit software environment. "While Verilog-A for compact models is relatively new, we made this decision only after hearing from multiple CMC members about successful use of this approach with multiple models," Watts said.
Regardless of where they stood on the PSP-vs.-HiSIM debate, vendors Silvaco included wasted little time in positioning themselves to capitalize on PSP's victory. Silvaco CEO Ivan Pesic said that while he believes that HiSIM-RF is the better model, Silvaco is ahead of competitors on PSP and plans to release a "perfect" PSP model this month.
The following day, Xpedion Design Systems Inc. announced a partnership with Gennady Gildenblat, the Penn State professor who co-developed the PSP model, to provide PSP parameter extraction for Xpedion modeling services. The company said it has already developed the first commercial PSP parameter-extraction technology through collaboration with Gildenblat.