There's a lot of excitement about the use of FinFETs at advanced process nodes, and no wonder, given their potential power and performance advantages over planar transistors. But CAD and methodology challenges remain, particularly when it comes to parasitic extraction. FinFET extraction challenges received a thorough review at the recent Electronic Design Process Symposium (EDPS) in a presentation by Tom Dillinger (right), CAD Technology Manager at Oracle.
Dillinger's talk was part of a "Foundry Day" on FinFETs at EDPS, a small but influential IEEE workshop that brings together some of the world's top experts in electronic design methodologies. The day was organized by Dan Nenni, founder and blogger at SemiWiki.com. Aparna Dey, senior technical marketing manager at Cadence, moderated the morning session that included Dillinger's talk. You can view the entire two-day EDPS program, and download presentation slides, here.
In an introductory talk, Nenni gave a brief overview of the FinFET value proposition, design challenges, and added complexities in terms of growing design rule manuals. Compared to planar transistors, performance is expected to increase 10-20% for the same power, or power is expected to decrease 25-40% at the same performance. Challenges include power density per unit, layout-dependent effects, new sources of process variation, new analog structures, and EDA tool challenges. "There are significant challenges, but the value proposition for FinFETs justifies the extra effort," Nenni said.
A Detailed Look at Extraction
Dillinger's presentation provided a FinFET transistor overview, discussed parasitic FinFET elements, and noted the challenges of extracting a FinFET parasitic model. Topics included the topology of the vertical fin and its impact on extraction; dummy gates and their impact; gate input capacitance and gate input equivalent resistance; advantages and shortcomings of the BSIM-CMG model; and the need to extract a reduced, equivalent electrical model.
Here are some of the takeaways from the presentation.
How FinFETs are made. Using a technique called "sidewall image transfer," a silicon fin is constructed through an anisotropic etching of a silicon layer, defined by a spacer deposited over a "sacrificial" patterned layer. Fins are always fabricated in pairs, so if an odd number of fins are required, a fin will need to be cut.
Fin profile is very important. Fins with a "rounded" profile are easier to fabricate, but introduce additional parasitic extraction challenges. Fin profile also affects leakage current substantially. Fins with very sharp, rectangular profiles lose some of the sub-threshold leakage savings promised by FinFET technology.
Sources of variation in FinFET geometry. These include fin height, fin thickness, fin corner rounding profile, fin sidewall roughness, gate line edge roughness, and gate CD variation over multiple parallel fins.
No body effect - good for digital, maybe not for analog. Since fins are fully depleted, there is very little body effect (Vt dependency on substrate bias). Digital designers see the body effect as detrimental and will welcome its disappearance. Some analog designers, however, make use of the body effect to tune transistor performance.
FinFET source and drain resistances are high. One cause is the use of a very narrow fin that is very lightly doped. To reduce Rs and Rd, an existing fabrication method called "selective epitaxial growth" can increase the volume of the source and drain regions.
BSIM-CMG model for FinFETs sets the standard. With BSIM-CMG, we have a standard model for FinFETs, and SPICE developers can start creating models. "We are full speed ahead on circuit simulation for FinFET technology."
But, BSIM-CMG model has some shortcomings. It uses an ideal single-fin model, which you simply multiply by the number of fins and fingers - this may be "a little deficient going forwards." BSIM-CMG model does not yet include layout-dependent effects.
Dummy gates are required for FinFET fabrication. This impacts parasitics, and makes it harder to migrate planar device layouts (which do not require dummy gates) into FinFET layouts.
New parasitic capacitances need extraction and modeling. 3D structures contribute to gate-to-source capacitance (Cgs), gate-to-drain capacitance (Cgd) and gate-to-substrate capacitance (Cgx). "I get great drive strength improvement, but I also have an increase in parasitic gate input capacitance as well. This is a big detraction for very high performance applications."
What model is needed for equivalent gate input resistance? The BSIM-CMG developers at U.C. Berkeley haven't accounted for gate resistance yet. "If this matters to you, FinFETs are still a work in progress."
A big issue from a design perspective is reduction. We have to reduce a tremendous amount of parasitic information into an equivalent electrical model that can be annotated to a designer's schematic. "Talk to your EDA vendor about extraction and parasitic reduction."
Parting comment: "I'm not worried as much that variation is going to blow up. The actual fin topology has a big impact on sub-threshold behavior, on parasitics, and on modeling. So you have to characterize it and give it to the technology file that the designers use when they abstract. That's the big one."
Slides from Dillinger's presentation are available here.