Architecture and CAD for nanoscale and three-dimensional FPGA.

摘要

FPGAs (field programmable gate arrays) are attractive alternatives compared to ASICs (application-specific integrated circuits) for significantly lowering amortized manufacturing costs and dramatically improving design productivity. The architecture of an FPGA is very regular. It is relatively easy to design a highly optimized tile, with consideration of various manufacturing related issues, and then to replicate it many times across the chip. The configurability of FPGAs also enables yield improvement and defect tolerance. However, FPGAs are still facing serious challenges in terms of delay, power consumption, and logic density compared to ASICs. FPGA is estimated to be over twenty times less efficient in logic density, over three times worse in delay, and over ten times higher in power consumption compared to a functionally equivalent ASIC.;One promising way to improve FPGA performance is to incorporate three-dimensional (3D) integration, which increases the number of active layers and optimizes the interconnect network vertically. Another solution is to apply novel nanoelectronic materials (nanomaterials) and devices. This dissertation introduces three novel reconfigurable architectures, named 3D nFPGA, FPCNA (field programmable carbon nanotube array), and NEM FPGA (nanoelectromechanical FPGA), which utilize 3D integration techniques and new nanoscale materials synergistically. Customized CAD flows that consider process variation have been developed for different architectures to evaluate their potential performances. Also described is a 3D variation aware routing flow, which is an essential tool for future 3D FPGA architecture exploration.;3D nFPGA is based on CMOS (complementary metal-oxide-semiconductor) and nano hybrid techniques that incorporate nanomaterials such as nanowire crossbars and carbon nanotube bundles into the CMOS fabrication process. Using unique features of FPGAs and a novel 3D stacking method enabled by the application of nanomaterials, 3D nFPGA obtains a 4× footprint reduction comparing to the traditional CMOS-based 2D FPGAs. The performance and power of 3D nFPGA driven by the 20 largest MCNC (microelectronics center of North Carolina) benchmarks have been evaluated. Results demonstrate that 3D nFPGA is able to provide a performance gain of 2.6× with a small power overhead compared to the traditional 2D FPGA.;FPCNA includes lookup tables created entirely from continuous carbon nanotube (CNT) ribbons. To determine the performance of the building blocks, variation aware physical design tools are used, with statistical static timing analysis (SSTA) that can handle both Gaussian and non-Gaussian random variables. A 2.75× performance improvement is seen over an equivalent CMOS FPGA at a 95% yield. In addition, FPCNA offers a 5× footprint reduction compared to a baseline FPGA.;3D NEM FPGA is the architecture that utilizes nanoelectromechanical (NEM) relays and 3D integration techniques synergistically. This proposed architecture has unique features including a hybrid CMOS-NEM FPGA lookup table (LUT) and configurable logic block (CLB), NEM-based switch block (SB) and connection block (CB), and face-to-face 3D stacking. This architecture also has a built-in feature called direct link, which takes advantage of the short vertical wire length provided by 3D stacking to further enhance performance. An overall 46.3% critical path delay reduction has been observed compared to its CMOS counterpart.;To maximize the potential performance gain of 3D integrated circuit architectures, an SSTA engine was developed to deal with both uncorrelated and correlated variations in 3D FPGAs. The effects of intra-die and inter-die variation are considered. Using the 3D physical design tool TPR as a base, a new 3D routing algorithm is developed, which improves the average performance of two-layer designs by over 22% and three-layer designs by over 27%.