DESIGN AND SYNTHESIS OF HIGH DENSITY INTEGRATED CIRCUITS

Gordon E. Moore, a co-founder of Fairchild Semiconductor, and later of Intel, predicted that after 1980 the complexity of an Integrated Circuit would be expected to double every two years. The prevision made by Moore held for decades, for this reason it is also called “Moore’s law”. The trend in ICs is driven by a reduction of area and power consumption. Today scaled CMOS technologies are the main solution for digital processing. However, the interconnection scaling is not optimal. At every new technology node, the number of metal layers and their thickness increases, exploiting the vertical direction. The reduction of the minimum distance between interconnections and the growth in vertical dimension increase the parasitic capacitance and consequently the dynamic power consumption. Moreover, due to the non-optimal scaling of the interconnections, signal routing is becoming more and more challenging at every technology node advancement. Very scaled technologies make possible to reach a great transistor density. However, the design must comply to strict rules for metal interconnections. The aim of this thesis is to find possible solutions to the disadvantages of scaled CMOS technologies. This goal is obtained in two different ways: using ad-hoc design techniques on today CMOS technologies and finding new approaches to logic synthesis of nanocrossbars, that are an emerging post-CMOS technology. The two approaches used corresponds to the two parts of this thesis. The first part presents the design of an Associative Memory focusing the attention on develop design and logic synthesis techniques to reduce power consumption. The field of applicability of AMs is real-time pattern-recognition tasks. The possible uses range from scientific calculations to image processing for intelligent autonomous devices to image reconstruction for electro-medical apparatuses. In particular AMs are used in High Energy Physics (HEP) experiments to detect particle tracks. HEP experiments generate a huge amount of data, but it is necessary to select and save only the most interesting tracks. Being the data compared in parallel, AMs are synchronous ICs that have a very peaked power consumption, and therefore it is necessary to minimize the power consumption. This AM is designed within the projects IMPART and HTT in 28 nm CMOS technology, using a fully-CMOS approach. The logic is based on the propagation of a “kill signal” that, if one of the bits in a word is not matching, inhibits the switching of the following cells. Thanks to this feature, the designed AM array consumes less than 0.7 fJ/bit. A prototype has been fabricated and it has proven to be functional. The final chip will be installed in the data acquisition chain of ATLAS experiment on HL-LHC at CERN. In the future nanocrossbars are expected to reduce device dimensions and interconnection complexity with respect to CMOS. Logic functions are obtained with switching lattices of four-terminal switches. The research activity on nanocrossbars is done within the project NANOxCOMP. To improve synthesis are used some algorithmic approaches based on Boolean function decomposition and regularities, in particular P-circuits, EXOR-Projected Sums of Products (EP-SOP), Dimension-reducible (D-red) functions and autosymmetric functions. The decomposed functions are implemented into lattices using internal and external decomposition methods. Experimental results show that this approaches reduce the complexity of the single synthesis problem and leads, in average, to a reduction of lattice area and synthesis time. Lattices are made of self-assembled structures and they have a non-negligible defectivity ratio. To cope with this limitation, some techniques to reduce sensitivity to defects have been studied.