Document Type



Doctor of Philosophy (PhD)


Electrical Engineering

First Advisor's Name

Arjuna Madanayake

First Advisor's Committee Title

Committee chair

Second Advisor's Name

S.I. Hariharan

Second Advisor's Committee Title

Committee member

Third Advisor's Name

Shubhendu Bhardwaj

Third Advisor's Committee Title

Committee member

Fourth Advisor's Name

Wujie Wen

Fourth Advisor's Committee Title

Committee member

Fifth Advisor's Name

Watson Lees

Fifth Advisor's Committee Title

Committee member


Analog computer, Continuous-time algorithms, PDE solvers, All-pass filters, Analog computing methods, Maxwell's equations, Wave equations, Nonlinear PDEs, CMOS analog computer, Stochastic calibration

Date of Defense



Analog computing (AC) was the predominant form of computing up to the end of World War II. The invention of digital computers (DCs) followed by developments in transistors and thereafter integrated circuits (IC), has led to exponential growth in DCs over the last few decades, making ACs a largely forgotten concept. However, as described by the impending slow-down of Moore’s law, the performance of DCs is no longer improving exponentially, as DCs are approaching clock speed, power dissipation, and transistor density limits. This research explores the possibility of employing AC concepts, albeit using modern IC technologies at radio frequency (RF) bandwidths, to obtain additional performance from existing IC platforms. Combining analog circuits with modern digital processors to perform arithmetic operations would make the computation potentially faster and more energy-efficient. Two AC techniques are explored for computing the approximate solutions of linear and nonlinear partial differential equations (PDEs), and they were verified by designing ACs for solving Maxwell's and wave equations. The designs were simulated in Cadence Spectre for different boundary conditions. The accuracies of the ACs were compared with finite-deference time-domain (FDTD) reference techniques.

The objective of this dissertation is to design software-defined ACs with complementary digital logic to perform approximate computations at speeds that are several orders of magnitude greater than competing methods. ACs trade accuracy of the computation for reduced power and increased throughput. Recent examples of ACs are accurate but have less than 25 kHz of analog bandwidth (Fcompute) for continuous-time (CT) operations. In this dissertation, a special-purpose AC, which has Fcompute = 30 MHz (an equivalent update rate of 625 MHz) at a power consumption of 200 mW, is presented. The proposed AC employes 180 nm CMOS technology and evaluates the approximate CT solution of the 1-D wave equation in space and time. The AC is 100x, 26x, 2.8x faster when compared to the MATLAB- and C-based FDTD solvers running on a computer, and systolic digital implementation of FDTD on a Xilinx RF-SoC ZCU1275 at 900 mW (x15 improvement in power-normalized performance compared to RF-SoC), respectively.




Files over 15MB may be slow to open. For best results, right-click and select "Save as..."



Rights Statement

Rights Statement

In Copyright. URI:
This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).