ABDELREHEEM, M. Návrh a vizualizace neuromorfních obvodů s využitím FPGA pro výpočetní techniku inspirovanou mozkem [online]. Brno: Vysoké učení technické v Brně. Fakulta elektrotechniky a komunikačních technologií. 2025.
This bachelor thesis focuses on the design, simulation, and FPGA-based implementation of neuromorphic circuits inspired by biological brain function. The student’s primary goal was to prototype a small-scale spiking neural network (SNN) on a Nexys A7-50T FPGA board using VHDL. The implementation includes digital Leaky Integrate-and-Fire (LIF) neurons, basic Spike-Timing Dependent Plasticity (STDP), and modules for spike routing, signal integration, and neuron interaction. From a technical perspective, the thesis demonstrates solid knowledge of neuromorphic principles and FPGA development. The student successfully designed a functional SNN of 2–5 neurons capable of basic processing tasks such as signal integration and pattern detection. He implemented modular components including clock_enable, spike_router, lif_neuron, and a top-level neuron_array, all tested via a structured VHDL testbench. The work is backed by waveform simulations, synthesis results, and RTL diagrams. One of the strongest aspects of this thesis is the author's ability to translate abstract biological mechanisms into a working hardware prototype. The design was verified through several test cases, including spike timing, weight adaptation, and input pattern variation. The author also performed a basic resource and power analysis using Vivado tools, showing reasonable utilization (e.g. ~3% LUTs and ~5% flip-flops), and identified opportunities for optimization, such as clock gating and weight refinement. Despite these accomplishments, several shortcomings must be mentioned. The project scope remained limited to a very small-scale network, and the STDP mechanism was basic, without advanced forms of dynamic learning or high-resolution weight adaptation. Additionally, although power consumption was analyzed (~102 mW), there is limited discussion of the scalability challenges or real-time integration with sensors, which are relevant for neuromorphic applications. Formally, the thesis is clearly structured and provides logical transitions between chapters. However, the language is occasionally too informal or journalistic in tone. There are stylistic inconsistencies, and some sections (especially early ones) include redundant explanations or imprecise grammar. The bibliography is relevant, but could be better formatted and more extensive given the current research activity in neuromorphic hardware. Considering that the student completed the entire project in approximately three months, the technical result is strong and shows a very good grasp of the domain. The limitations mostly lie in the scope, depth of hardware optimization, and professional presentation.
This thesis deals with the description and implementation of neuromorphic circuits, i.e. the realisation of a new class of artificially created structures inspired by the functioning of the brain. In the opening sections, author compares the approach based on digital logic with neuromorphic circuits, which have the potential to overcome the computing power of current technology, considering the physical and technical limitations of current digital circuits. From this perspective, the work is very interesting and relevant. Author chose a rather challenging topic and successfully implemented biologically inspired mechanisms into the hardware design. He then tested the resulting prototype neural network in a VHDL environment. In conclusion, he compared the results achieved with state-of-the-art neuromorphic systems from IBM and INTEL. The text is 62 pages long, which slightly exceeds the recommended length for a bachelor's thesis. In terms of presentation, the thesis is of a very good standard, with the text well structured and illustrated. From a formal point of view, the text also meets the requirements for a bachelor's thesis, although it is sometimes grammatically inaccurate. However, the conclusion is very brief and I think it could be elaborated on a little more. I consider the assignment to be fulfilled; the student has demonstrated very good creative abilities by implementing the above-described neural network into a functional prototype. In view of the above, I rate the thesis with 85 points.
eVSKP id 169691