Shortening the gap between pre- and post-layout analog IC performance by reducing the LDE-induced variations with multi-objective simulated quantum annealing
Martins, R. M.
Póvoa , R. P.
Engineering Applications of Artificial Intelligence Vol. 98, Nº n/a, pp. 104102 - 104102, February, 2021.
ISSN (print): 0952-1976
Scimago Journal Ranking: 1,11 (in 2020)
Digital Object Identifier: 10.1016/j.engappai.2020.104102
The design of analog and mixed-signal integrated circuits (ICs) is intricate due to the continuous nature of the signals handled. Still, it is also strongly affected by the physical implementation of analog devices on the circuits’ layout. The circuit layout corresponds to the physical implementation of an analog IC used in fabrication that describes its devices geometrically. As circuits’ integration and device sizes shrink, the physics of the interactions between devices, as they are placed in the layout, was proved to easily drive analog and mixed-signal ICs from promising pre-layout performances to completely post-layout malfunction. As these layout-dependent effects (LDEs) can only be evaluated once the layout is completed, the true post-layout performance is only evaluated in a late stage of the traditional design flow, causing expensive redesign iterations lacking the information that identifies precisely where, in the layout, there are problems needed to be solved. For technologies above the 40-nanometers, the leading causes of LDEs are mobility and threshold voltage variations. This paper proposes an automatic device placement methodology that explicitly accounts for, and minimizes, these LDEs. An absolute representation of the floorplan is adopted, and, multiple optimization techniques, including the novel, constrained archive-based multi-objective implementation of the simulated quantum annealing inspired algorithm, enhanced with specific LDE-impact mitigation operators are applied to solve the problem. In each of these optimization processes, established LDE formulations for accurate circuit simulation models are used to evaluate each candidate placement solution, and, guide the optimization process. In the case of multi-objective implementations, ultimately offering a realistic perspective of the LDE-aware design tradeoffs between performance deterioration and used chip area. Experimental results conducted over state-of-the-art analog structures on a challenging 65-nanometers technology node show that the proposed methodology shortens the gap between pre- and post-layout performance by reducing the LDE-induced variations, aiming for first-time-right layout design.