Donald Bren School of Information and Computer Sciences
An approach to component generation and technology adaptation
- Author(s): Kipps, James Randall
- et al.
Component generation is the task of mapping the abstract functional specification of register-transfer (RT) components, such as decoders and multiplexers, adders and comparators, and multipliers and arithmetic logic units, into configurations of connected physical layout cells. Cells are drawn from a given ASIC (application-specific integrated circuit) library.
In this dissertation, I describe a symbolic pattern-matching approach to component generation and, relative to this, an approach to automating technology adaptation. I define the component decomposition algorithm and technology compilation algorithm that formalize these two approaches and describe implementations of each, in the DTAS component generation system and the LOLA technology adaptation system, respectively. I present empirical results to validate the utility of my approach to component generation, and I present a demonstration to validate my approach to technology adaptation.
My approach to component generation has two significant benefits. First, it enables the use of complex functional library cells, such as adders and CLAs, in the generation of designs for functional units. Second, it effectively searches the design space for designs that make desirable tradeoffs between design constraints, such as area and delay. My approach to technology adaptation is significant because it bootstraps the DTAS component generation system into new ASIC cell libraries, as well as cell libraries as they undergo change. In this way, the technology compilation algorithm automates the task of maintaining technology independence.
To validate my approach to component generation, I present the results of four sets of experiments using the DTAS component generation system. The first set examines the effectiveness of search control in DTAS; the second examines the capability to find desirable design alternatives; the third compares designs generated by DTAS with those of MISII; and the fourth shows how the use of complex library cells improves design quality. To validate my approach to automating technology adaptation, I demonstrate the application of the LOLA technology adaptation system to a cell library as it undergoes four phases of evolution.