Electrical Engineering
      and Computer Sciences

Electrical Engineering and Computer Sciences

COLLEGE OF ENGINEERING

UC Berkeley

Evolutionary MOSFET Structure and Channel Design for Nanoscale CMOS Technology

Byron Ho

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2012-213
November 21, 2012

http://www.eecs.berkeley.edu/Pubs/TechRpts/2012/EECS-2012-213.pdf

The constant pace of CMOS technology scaling has enabled continuous improvement in integrated-circuit cost and functionality, generating a new paradigm shift towards mobile computing. However, as the MOSFET dimensions are scaled below 30nm, electrostatic integrity and device variability become harder to control, degrading circuit performance. In order to overcome these issues, device engineers have started transitioning from the conventional planar bulk MOSFET toward revolutionary thin-body transistor structures such as the FinFET or fully-depleted silicon-on-insulator (FDSOI) MOSFET. While these alternatives appear to be elegant solutions, they require increased process complexity and/or more expensive starting substrates, making development and manufacturing costs a concern. For certain applications (such as mobile electronics), cost is still an important factor, inhibiting the quick adoption of the FinFET and FDSOI MOSFET structures while providing an opportunity to extend the competitiveness of planar bulk-silicon CMOS. A segmented-channel MOSFET (SegFET) design, which combines the benefits of both planar bulk MOSFETs (i.e. lower process complexity and/or cost) and thin-body transistor structures (i.e. improved electrostatic integrity), can provide an evolutionary pathway to enable the continued scaling of planar bulk technology below 20nm. In this work, experimental results comparing SegFETs and planar MOSFETs show suppressed short-channel effects and comparable on-state current (despite halving the effective device width). In addition, three-dimensional device simulations were used to optimize and benchmark the bulk SegFET and FinFET designs. Compared to the FinFET design, the results indicate that the SegFET can achieve similar on-state current performance and intrinsic delay (for the same channel stripe pitch) at a lower height/width aspect ratio and less aggressive retrograde channel doping gradient for improved manufacturability, making it a promising candidate for continued bulk-silicon CMOS transistor scaling. High-mobility channels are also investigated in this work for their potential to improve MOSFET performance, but issues with physical material parameters (electrostatic control, strain effects, etc.) and process integration necessitate careful design when implementing these materials in the MOSFET channel regions. Because germanium (Ge) and silicon-germanium (Si1-xGex) alloys are Group IV materials like silicon (Si), and since these materials are already extensively used in mainstream volume integrated-circuit manufacturing, they represent the most straightforward path to integrating high-mobility channels on silicon. Device simulations are used to optimize Si1-xGex channel thickness and Ge concentration for Si1-xGex/Si heterostructure p-channel MOSFETs; it is found that a thin (< 5 nm) channel with moderate (20% - 40%) Ge concentration is optimal for device performance and manufacturability. Si1-xGex/Si heterostructure channels were also experimentally integrated in SegFETs and show 30% higher on-state current (for an off-state current of 10 nA/µm) and reduced layout-width dependencies as compared to the planar MOSFET counterpart. Finally, Monte Carlo simulations were used to compare the scalability and performance of pure Ge and Si double-gate structures at short gate lengths (< 20 nm). Due to the higher dielectric constant and low transport mass (which becomes lighter with added strain), pure Ge channels may not be attractive for ultra-short gate lengths (8 nm) because of reduced electrostatic control and increased direct source-to-drain tunneling. However, if gate length scaling slows dramatically or channel strain decreases with shrinking device pitch, then Ge channels can be a viable alternative to Si for high-performance applications.

Advisor: Tsu-Jae King Liu


BibTeX citation:

@phdthesis{Ho:EECS-2012-213,
    Author = {Ho, Byron},
    Title = {Evolutionary MOSFET Structure and Channel Design for Nanoscale CMOS Technology},
    School = {EECS Department, University of California, Berkeley},
    Year = {2012},
    Month = {Nov},
    URL = {http://www.eecs.berkeley.edu/Pubs/TechRpts/2012/EECS-2012-213.html},
    Number = {UCB/EECS-2012-213},
    Abstract = {        The constant pace of CMOS technology scaling has enabled continuous improvement in integrated-circuit cost and functionality, generating a new paradigm shift towards mobile computing.  However, as the MOSFET dimensions are scaled below 30nm, electrostatic integrity and device variability become harder to control, degrading circuit performance.  In order to overcome these issues, device engineers have started transitioning from the conventional planar bulk MOSFET toward revolutionary thin-body transistor structures such as the FinFET or fully-depleted silicon-on-insulator (FDSOI) MOSFET.  While these alternatives appear to be elegant solutions, they require increased process complexity and/or more expensive starting substrates, making development and manufacturing costs a concern.  
	For certain applications (such as mobile electronics), cost is still an important factor, inhibiting the quick adoption of the FinFET and FDSOI MOSFET structures while providing an opportunity to extend the competitiveness of planar bulk-silicon CMOS.  A segmented-channel MOSFET (SegFET) design, which combines the benefits of both planar bulk MOSFETs (i.e. lower process complexity and/or cost) and thin-body transistor structures (i.e. improved electrostatic integrity), can provide an evolutionary pathway to enable the continued scaling of planar bulk technology below 20nm.  In this work, experimental results comparing SegFETs and planar MOSFETs show suppressed short-channel effects and comparable on-state current (despite halving the effective device width).  In addition, three-dimensional device simulations were used to optimize and benchmark the bulk SegFET and FinFET designs.  Compared to the FinFET design, the results indicate that the SegFET can achieve similar on-state current performance and intrinsic delay (for the same channel stripe pitch) at a lower height/width aspect ratio and less aggressive retrograde channel doping gradient for improved manufacturability, making it a promising candidate for continued bulk-silicon CMOS transistor scaling.
	High-mobility channels are also investigated in this work for their potential to improve MOSFET performance, but issues with physical material parameters (electrostatic control, strain effects, etc.) and process integration necessitate careful design when implementing these materials in the MOSFET channel regions.  Because germanium (Ge) and silicon-germanium (Si1-xGex) alloys are Group IV materials like silicon (Si), and since these materials are already extensively used in mainstream volume integrated-circuit manufacturing, they represent the most straightforward path to integrating high-mobility channels on silicon.  Device simulations are used to optimize Si1-xGex channel thickness and Ge concentration for Si1-xGex/Si heterostructure p-channel MOSFETs; it is found that a thin (< 5 nm) channel with moderate (20% - 40%) Ge concentration is optimal for device performance and manufacturability.  Si1-xGex/Si heterostructure channels were also experimentally integrated in SegFETs and show 30% higher on-state current (for an off-state current of 10 nA/µm) and reduced layout-width dependencies as compared to the planar MOSFET counterpart.  Finally, Monte Carlo simulations were used to compare the scalability and performance of pure Ge and Si double-gate structures at short gate lengths (< 20 nm).  Due to the higher dielectric constant and low transport mass (which becomes lighter with added strain), pure Ge channels may not be attractive for ultra-short gate lengths (8 nm) because of reduced electrostatic control and increased direct source-to-drain tunneling.  However, if gate length scaling slows dramatically or channel strain decreases with shrinking device pitch, then Ge channels can be a viable alternative to Si for high-performance applications.}
}

EndNote citation:

%0 Thesis
%A Ho, Byron
%T Evolutionary MOSFET Structure and Channel Design for Nanoscale CMOS Technology
%I EECS Department, University of California, Berkeley
%D 2012
%8 November 21
%@ UCB/EECS-2012-213
%U http://www.eecs.berkeley.edu/Pubs/TechRpts/2012/EECS-2012-213.html
%F Ho:EECS-2012-213