Comparative Evaluation of Deep Neural Network Implementations Across Different Neural Processing Units

Type of work:

Master Thesis

Assignment:

The objective of this master thesis is to investigate and compare the implementation efficiency of selected DNN models across different NPU platforms. The student will analyze how architectural characteristics and software toolchains influence performance, resource utilization, latency, throughput, and energy efficiency.

Skills:

• Good programming skills (e.g., Python, C/C++)
• Basic knowledge of machine learning and deep neural networks
• Interest in embedded systems, hardware accelerators, and AI systems

Background:

Deep Neural Networks (DNNs) have become a fundamental technology in modern artificial intelligence applications, ranging from computer vision and natural language processing to autonomous systems and edge computing. To meet the increasing computational demands of DNN inference and training, specialized hardware accelerators known as Neural Processing Units (NPUs) have emerged. These platforms differ significantly in terms of architecture, supported software stacks, memory hierarchy, power efficiency, and 
 

Supervisor:

Dr.-Ing. Vladimir Rybalkin

Student:

Leon Gualdron Daniel Felipe

Year:

2026