Ilknur Mustafazade

Ilknur Mustafazade

Auto-generated profile, based on publicly available information

Ilknur Mustafazade

Drexel University, Philadelphia, USA

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
23 September 2024 in IEEE Embedded Systems Letters
Verified
Journal Article
Closed access
PHPhu Khanh Huynh
Phu Khanh Huynh
IMIlknur Mustafazade
Ilknur Mustafazade
FCFrancky CATTHOOR
Francky CATTHOOR
NKNagarajan Kandasamy
Nagarajan Kandasamy
ADAnup Das
Anup Das
Large-scale neuromorphic architectures consist of computing tiles that communicate spikes using a shared interconnect. We propose ADIONA, a dynamic segmented bus interconnect to address design scalability while reducing energy and latency of spike traffic. A DIONA c onsists o f p arallel bus lanes arranged in a ladder-shaped structure that allows any tile to connect to another, offers multiple routing options for communication links, and provides a high level of customization for different mapping scenarios and use cases. Each lane in the ladder bus is partitioned into segments using lightweight buffer-less switches. Based on compile-time communication information, these switches can be dynamically reconfigured at runtime to execute the target application. Our dynamic segmented bus interconnect substantially enhances hardware utilization, improves fault tolerance, and offers adaptability to execute different applications on a single hardware platform. We evaluate ADIONA using three synthetic and three realistic machine learning workloads on a cycle-accurate neuromorphic simulator. Our results show that ADIONA reduces energy consumption by 2.1×, latency by 40×, and interconnect area by 2×, compared to a state-of-the-art interconnect for neuromorphic systems.


7 May 2024
Verified
Conference Paper
Open access
IMIlknur Mustafazade
Ilknur Mustafazade
NKNagarajan Kandasamy
Nagarajan Kandasamy
ADAnup Das
Anup Das
Neuromorphic hardware, designed to mimic the neural structure of the human brain, offers an energy-efficient platform for implementing machine-learning models in the form of Spiking Neural Networks (SNNs). Achieving efficient SNN execution on this hardware requires careful consideration of various objectives, such as optimizing utilization of individual neuromorphic cores and minimizing inter-core communication. Unlike previous approaches that overlooked the architecture of the neuromorphic core when clustering the SNN into smaller networks, our approach uses architecture-aware algorithms to ensure that the resulting clusters can be effectively mapped to the core. We base our approach on a crossbar architecture for each neuromorphic core. We start with a basic architecture where neurons can only be mapped to the columns of the crossbar. Our technique partitions the SNN into clusters of neurons and synapses, ensuring that each cluster fits within the crossbar's confines, and when multiple clusters are allocated to a single crossbar, we maximize resource utilization by efficiently reusing crossbar resources. We then expand this technique to accommodate an enhanced architecture that allows neurons to be mapped not only to the crossbar's columns but also to its rows, with the aim of further optimizing utilization. To evaluate the performance of these techniques, assuming a multi-core neuromorphic architecture, we assess factors such as the number of crossbars used and the average crossbar utilization. Our evaluation includes both synthetically generated SNNs and spiking versions of well-known machine-learning models: LeNet, AlexNet, DenseNet, and ResNet. We also investigate how the structure of the SNN impacts solution quality and discuss approaches to improve it.

12 September 2025
Verified
Thesis
Closed access
IMIlknur Mustafazade
Ilknur Mustafazade
NKNagarajan Kandasamy
Nagarajan Kandasamy
ADAnup Das
Anup Das
Neuromorphic computing systems implementing spiking neural networks (SNNs) offer significant advantages in energy efficiency and real-time processing for brain-inspired applications. However, deploying large-scale SNNs on neuromorphic hardware faces critical challenges in resource allocation and reliability assurance that limit their practical adoption. Conventional clustering approaches, while effective for traditional computing architectures, fail to account for the unique constraints of neuromorphic cores such as crossbar capacity limitations and synaptic mapping requirements. Moreover, the deployment of neuromorphic systems in mission-critical applications necessitates robust fault detection capabilities that current approaches do not adequately address. This dissertation presents a comprehensive framework for efficient and reliable SNN deployment on neuromorphic hardware through architecture-aware clustering and integrated fault detection mechanisms. The first major contribution is the development of architecture-aware clustering algorithms that explicitly consider crossbar constraints during SNN partitioning. Beginning with the foundational crossbar architecture, we demonstrate how synaptic-connection-aware clustering significantly improves resource utilization compared to traditional graph partitioning methods, resulting in 1.49-1.83x better resource utilization for crossbars. Building upon this foundation, the ATLAS framework addresses the performance speed and solution quality by employing specialized data structures and meta-heuristic optimizations. Our work resulted in dramatic computational improvements by reducing clustering time from over 24 hours to under 20 minutes for large networks with substantial energy consumption reductions from 12.5% to 45.7% across evaluated models. As part of this work, we also introduce a software framework that abstracts architecture-specific constraints, enabling portability across various neuromorphic backends. The next major contribution introduces a software-based built-in self-test (BIST) strategy that leverages alarm placement theory to minimize fault detection overhead while ensuring comprehensive coverage. The alarm placement algorithm is applied to SNNs trained on various datasets resulting in high test coverage across diverse fault models such as stuck-at faults and resistance drifts. Multiple methods are introduced to enable the application of alarm placement on much larger workloads. Through channel-based abstraction and trace-based analysis, we achieve complexity reductions of 32x for LeNet and 248x for AlexNet while maintaining effective fault coverage exceeding 96-98%. Together, these contributions establish a complete methodology for neuromorphic SNN deployment that addresses both efficiency and reliability requirements, providing a foundation for practical large-scale neuromorphic computing systems in safety-critical applications.