Article: Mathematical processing units: a prototype for enabling next-generation ambient AI Journal: International Journal of Critical Computer-Based Systems (IJCCBS) 2026 Vol.12 No.5 pp.1 - 21 Abstract: The next generation of AI spanning from large language models (LLMs) to joint embedding predictive architectures (JEPAs) depend heavily on high-quality ambient data acquisition. However, the effectiveness of these future models is directly tied to the fidelity and efficiency of their training data pipelines. We present a proof-of-concept for a new type of processing hardware called mathematical processing units (MPUs), built upon commercially available FPGAs and designed to acquire and pre-process ambient sensor data with high fidelity, low power, and low cost, while also aiding in sensor fusion and fault tolerance. We demonstrate the effectiveness of the proposed approach by performing Kalman filtering on environmental data. Across six sensing scenarios, the MPU-based Kalman filter implementation reduced measurement noise by an average of 77.2%, demonstrating substantial improvement in signal fidelity. These results highlight the potential of MPUs as core infrastructure for future AI systems that interact with the physical world. Inderscience Publishers - linking academia, business and industry through research

Title: Mathematical processing units: a prototype for enabling next-generation ambient AI

Authors: Rohan Sinha; Aadi Jain; Dev Syam; Shivam Kundan

Addresses: Woodside Priory School, 302 Portola Rd., Portola Valley, CA 94028, USA ' School of Computer Science, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, PA 15213, USA ' GEMS Modern Academy, Nad al Sheba, Dubai, United Arab Emirates ' School of Electrical, Computer and Biomedical Engineering, Southern Illinois University, 1230 Lincoln Drive, Carbondale, IL 62901, USA

Abstract: The next generation of AI spanning from large language models (LLMs) to joint embedding predictive architectures (JEPAs) depend heavily on high-quality ambient data acquisition. However, the effectiveness of these future models is directly tied to the fidelity and efficiency of their training data pipelines. We present a proof-of-concept for a new type of processing hardware called mathematical processing units (MPUs), built upon commercially available FPGAs and designed to acquire and pre-process ambient sensor data with high fidelity, low power, and low cost, while also aiding in sensor fusion and fault tolerance. We demonstrate the effectiveness of the proposed approach by performing Kalman filtering on environmental data. Across six sensing scenarios, the MPU-based Kalman filter implementation reduced measurement noise by an average of 77.2%, demonstrating substantial improvement in signal fidelity. These results highlight the potential of MPUs as core infrastructure for future AI systems that interact with the physical world.

Keywords: FPGA; Kalman filtering; ambient intelligence; signal processing; VLSI design; large language models; LLMs; mathematical processing units; MPUs.

DOI: 10.1504/IJCCBS.2026.157298

International Journal of Critical Computer-Based Systems, 2026 Vol.12 No.5, pp.1 - 21

Received: 02 Jul 2025
Accepted: 07 Aug 2026
Published online: 27 Oct 2026 *

Title: Mathematical processing units: a prototype for enabling next-generation ambient AI

Keep up-to-date