doc. Ing. Jan Schmidt, Ph.D.

Contemporary industrial design and verification of digital systems has strong demands on time-to-market and hence design time. Often, we hear requirements such as “1 percent better result for 1 percent longer computation”. In this situation, accelerating the computation seems desirable. Yet, current algorithms have been designed with sequential processing in mind. They usually are complex, multi-layer iterative heuristics. To accelerate such algorithms, it is necessary either find parallelism e.g., in branch-and-bound heuristics, or replace some parts of the algorithm with better parallelizable algorithms. Further, we need to investigate which architectures and what granularity suit the designed algorithm best. It is necessary to keep work-optimality in mind, as it projects itself directly into energy efficiency.

Approximate computing is a collection of design methods for systems that fulfill their purposes but have dramatically lower demands for implementation resources, such as power or area. There are two metrics for the approximation in use. The first one is arithmetic difference and is used whenever a set of binary values is interpreted as a number. The second one is the Hamming distance, that is, the number of binary values differing from original specification. There is a number of methods for both of the metrics in development. However, the true meaning of the approximation metric should follow the intended purpose of the system. In the case of image representation, arithmetic difference usually suffices. In other applications, besides varying metrics, also hard constraints appear, that is, constraints that the approximation may not violate. Most of contemporary methods cannot accept such constraints and metrics. The aim of the work is to develop methods able to accept application-specific constraints and metrics, and, eventually, document the price that must be paid for such a versatility.

doc. Ing. Jan Schmidt, Ph.D.

Ing. Tomáš Zahradnický, Ph.D.

Department of Digital Design

This work focuses on the design and implementation of an assembler, emulator and a debugging application for an instruction set of a simple processor, which will enable beginners to orient themselves in the field of programming in the assembly language. It allows constraining of the instruction set for exercises, in which we want to demonstrate specific attributes of assembly programming.

Thesis on DSpace

doc. Ing. Jan Schmidt, Ph.D.

Ing. Michal Prokš

Department of Digital Design

doc. Ing. Jan Schmidt, Ph.D.

Ing. Stanislav Trojan

Department of Digital Design

doc. Ing. Jan Schmidt, Ph.D.

Ing. Jiří Dostál, Ph.D.

Department of Digital Design

doc. Ing. Jan Schmidt, Ph.D.

Dr.-Ing. Martin Novotný

Department of Digital Design