Peter Forcha

Mr. Peter Forcha

orcid

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

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

Representative Publications

View All Publications
10 March 2026 in Electronics
Verified
Journal Article
Open access
PDF file
PMPeter Esenju Mbua
Peter Esenju Mbua
PFPeter Forcha
Peter Forcha
Christophe BobdaChristophe Bobda
Christophe Bobda
The emergence of chiplet-based architectures represents a paradigm shift in post-Moore’s Law computing systems, offering substantial cost and yield advantages through functional disaggregation. However, the heterogeneity of inter-chiplet communication introduces unique performance challenges that conventional partitioning strategies fail to address. In this work, the ways in which poor workload partitioning degrades communication performance in chiplet-based systems are comprehensively characterized. We demonstrate, through a detailed experimental analysis, that suboptimal workload partitioning can increase inter-chiplet communication latency by up to a factor of 10 and inflate network congestion beyond sustainable levels as systems scale. Our findings show that optimized partitioning strategies can achieve an 87.4% reduction in inter-chiplet traffic, improve system throughput by a factor of 8.75, and enhance energy efficiency by a factor of 10.3 compared to naive partitioning approaches. We further characterize how these effects scale with system size, revealing that the communication overhead can consume 85% of the execution time in poorly partitioned 16-chiplet systems, compared to only 35% in well-partitioned configurations. This work provides essential insights into the communication-aware design space of chiplet systems and validates the critical importance of sophisticated workload partitioning algorithms.

27 February 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
PMPeter Esenju Mbua
Peter Esenju Mbua
PFPeter Forcha
Peter Forcha
Christophe BobdaChristophe Bobda
Christophe Bobda
The emergence of chiplet-based architectures represents a paradigm shift in post-Moore’s Law computing systems, offering substantial cost and yield advantages through functional disaggregation. However, the heterogeneity of inter-chiplet communication introduces unique performance challenges that conventional partitioning strategies fail to address. This work presents a comprehensive characterization of how poor workload partitioning degrades communication performance in chiplet-based systems. We demonstrate, through detailed experimental analysis, that suboptimal workload partitioning can increase inter-chiplet communication latency by up to 10×, and can inflate network congestion beyond sustainable levels as systems scale. Our findings show that optimized partitioning strategies can achieve 87.4% reduction in inter-chiplet traffic, improve system throughput by 8.75×, and enhance energy efficiency by 10.3× compared to naive partitioning approaches. We further characterize how these effects compound with system scalability, revealing that communication overhead can consume 85% of execution time in poorly partitioned 16-chiplet systems, versus only 35% in well partitioned configurations. This work provides essential insights into the communication-aware design space of chiplet systems and validates the critical importance of sophisticated workload partitioning algorithms.

5 February 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
MPMaximillian Panoff
Maximillian Panoff
AHAntonio Hendricks
Antonio Hendricks
PFPeter Forcha
Peter Forcha
JMJung Minchul
Jung Minchul
SWSHUO WANG
SHUO WANG
Christophe BobdaChristophe Bobda
Christophe Bobda
As robots and other forms of intelligent automation are improved, solutions to safely and accurately control their operation have also evolved. Many recent solutions to this control problem rely heavily upon Digital Twins (DT), or a digital copy of the environment, which tracks the presence and location of all agents and objects. However, implementing a digital twin is a non-neglible challenge as a high fidelity link between the real and digital worlds must be maintained. To this end, we have compiled a brief background on the problem of 3-Dimensional (3D) Pose Recovery from markerless visual data, listing some of the most recent works in the field and comparing their suitability for DT. In 3D Pose Recovery (3DPR), a subject’s position and orientation, or pose, is recovered from visual data in which objects and actors are not physically tagged or identified. This pose can then be used to help identify the location of a target object or actor to track in the digital twin. Additionally, we identify some emerging or remaining challenges in visual 3DPR in the context of DT and highlight a few methods that attempt to address them.

2 February 2025 in Technologies
Verified
Journal Article
Open access
PDF file
MPMaximillian Panoff
Maximillian Panoff
JAJoshua Acevedo
Joshua Acevedo
HYHonggang Yu
Honggang Yu
PFPeter Forcha
Peter Forcha
SWSHUO WANG
SHUO WANG
Christophe BobdaChristophe Bobda
Christophe Bobda
Determining human intention is a challenging task. Many existing techniques seek to address it by combining many forms of data, such as images, point clouds, poses, and others, creating multi-modal models. However, these techniques still often require significant foreknowledge in the form of known potential activities and objects in the environment, as well as specific types of data to collect. To address these limitations, we propose Intent-BERT and Universal Context Encoders, which combine to form workload-agnostic framework that can be used to predict the next activity that a human performs as an Open Vocabulary Problem and the time until that switch, along with the time the current activity ends. Universal Context Encoders utilize the distances between the embeddings of words to extract relationships between Human-Readable English descriptions of both the current task and the origin of various multi-modal inputs to determine how to weigh the values themselves. We examine the effectiveness of this approach by creating a multi-modal model using it and training it on the InHARD dataset. It is able to return a completely accurate description of the next Action performed by a human working alongside a robot in a manufacturing task in ∼42% of test cases and has a 95% Top-3 accuracy, all from a single time point, outperforming multi-modal gpt4o by about 50% on a token by token basis.