To improve lives by enabling personalized AI experiences on battery-powered devices, ensuring unparalleled performance without compromising user privacy and security.
By developing revolutionary semiconductor IP, we are creating the computational infrastructure that makes intelligent, context-aware personal technology not just possible but seamless and trustworthy.
The NSF SBIR (Small Business Innovation Research) program is a competitive federal grant that funds early-stage R&D at U.S. startups developing high-risk, novel, innovative technologies. It provides non-dilutive funding (up to $1.7M across phases) to projects with strong scientific merit and commercial potential. Only about 10-15% of applicants receive awards, making it a selective validation of a startup’s technology and market viability.
AFWERX
STTR from USAF
The SBIR AFWERX Award is a prestigious grant from the U.S. Air Force, in partnership with AFWERX and the Department of Defense, designed to fund innovative startups working on cutting-edge defense and dual-use technologies. What makes it special is its highly competitive nature—only the most groundbreaking solutions with real-world impact are selected, giving our startup credibility and validation in the eyes of customers. By earning this award, we’ve proven that our technology meets the highest standards of innovation, reliability, and mission readiness.
The SBIR AFWERX Award is a prestigious grant from the U.S. Air Force, in partnership with AFWERX and the Department of Defense, designed to fund innovative startups working on cutting-edge defense and dual-use technologies. What makes it special is its highly competitive nature—only the most groundbreaking solutions with real-world impact are selected, giving our startup credibility and validation in the eyes of customers. By earning this award, we’ve proven that our technology meets the highest standards of innovation, reliability, and mission readiness, ensuring you’re investing in a proven, future-ready solution.
Partnering with AFWERX and the Department of Defense, this highly selective grant is reserved for technologies that meet rigorous standards for innovation and real-world impact. By earning this award, we’ve demonstrated that our solutions deliver the reliability and performance demanded by mission-critical defense applications—while remaining adaptable for commercial markets.
Recognition From…
Statement Of Capabilities
Arith Inc. was founded by a distinguished co-author of the IEEE 754 Standard for Floating-Point Arithmetic. Our pioneering, high-performance, and energy-efficient silicon intellectual property (IP) has enabled breakthrough technologies—including the world’s first augmented reality (AR) glasses featuring contextual artificial intelligence (AI). mightyFP™, our flagship AI acceleration IP, delivers a rich suite of arithmetic and AI compute capabilities essential for next-generation edge and data center applications. These include: Precision conversions with rounding between FP32/16, BF16, FP8/6/4, and integer formats Quantized conversions between fixed-point data of arbitrary formats Mixed-type and mixed-precision operations, such as addition, subtraction, summation, sorting, multiplication, FMA, and (M)MAC End-to-end AI compute acceleration, spanning dot products, matrix multiplications, convolutions, and activation functions such as ReLU, GELU, sigmoid, tanh, erf, and fast exponentials optimized for attention mechanisms FFT and trigonometric function optimizations, including parallel add/subtract, complex multipliers, sine, cosine, tangent, and arctangent Mathematical functions, including exponentiation, logarithms (base 2, e, and 10), reciprocal, square root, and reciprocal square root These capabilities are designed for ultra-low-latency, single-digit cycle performance, enabling energy-efficient AI inference, real-time signal processing, and scalable deployment across diverse mission environments
Products
MightyTP(TM)
mightyFP™
A silicon-proven AI acceleration IP core designed to maximize compute efficiency in power- and latency-constrained environments. Purpose-built for both edge and datacenter deployment, it delivers unmatched performance-per-watt by fusing arithmetic, logic, and AI processing into a compact, modular, highly parallel architecture. With modular configurability and broad compatibility across ASICs, FPGAs, and custom SoCs, mightyFP™ empowers chip and system designers to deploy real-time intelligence at scale—whether in autonomous platforms, secure datacenters, mission-critical devices, or consumer products. Fully customizable, and manufacturing-process agnostic, mightyFP™ enables the next wave of intelligent systems without compromising power, size, or control.
Success Stories
Datacenter Deployment — Accelerating LLM Inference at Scale
A leading platform company integrated a version of mightyFP™ into their AI inference pipeline to address the rising costs and energy demands of serving large language models (LLMs). Facing power limitations and server rack density challenges, the customer replaced several GPU-based inference nodes with mightyFP™-enabled ASICs. By leveraging mightyFP™’s support for ultra-low precision formats (e.g., FP8), the team achieved a 3× improvement in inference throughput and a 65% reduction in power consumption, all while maintaining model accuracy. The drop-in IP block allowed rapid deployment without modifying the rest of their compute stack. As a result, the company slashed cost-per-query, reduced cooling infrastructure needs, and improved sustainability metrics—gaining a competitive edge in AI infrastructure.
AR/VR/XR Deployment — Enabling On-Device Intelligence for Smart Glasses
An enterprise wearable manufacturer adopted mightyFP™ to power a new generation of context-aware AR glasses. The challenge was clear: perform real-time computer vision, and environment-aware overlays—entirely on-device, with no cloud dependency and minimal heat dissipation. By integrating mightyFP™ into their custom SoC, the company enabled AI-driven features such as object identification, and hazard detection. mightyFP™’s compact data formats allowed high-speed, low-latency inference without draining the battery or overheating the form factor. The glasses extended operational time by 40%, passed ruggedization tests.
TEAM
David Chen
CEO/Founder
David Chen is a seasoned professional with over 30 years of experience in engineering, management, standardization, and commercialization in the semiconductor industry. He co-authored IEEE Std 754 and serves as a voting member on the IEEE 3109 committee. As the sole developer of Cadence®/Tensilica® Floating-Point Units (PIF), he has demonstrated technical expertise. Chen co-founded Alayar’es, a Silicon Valley startup, and holds multiple U.S. patents in Applied AI, Floating-Point Algorithms, Computer Arithmetic, and the CPII Model..
Prof. James Demmel
Researcher for Development
James Demmel, a distinguished professor at UC Berkeley and faculty member at Lawrence Berkeley Lab, serves as TPOC of the UC Berkeley Research Institute. A renowned expert in numerical algorithms, he is a Fellow of ACM, IEEE, SIAM, and AMS, and a member of the National Academy of Engineering. Demmel is partnering with Artth to develop a simulation package for MIGHTYFP™, bringing his deep expertise in matrix multiplication and communication-avoiding algorithms. He is best known for his foundational contributions to LAPACK, the industry-standard library for numerical linear algebra.
William Huffman
Consultant
William Huffman brings over 40 years of engineering and management expertise in the semiconductor industry. He plays a key role in ensuring MIGHTYFP™’s compatibility with diverse processor architectures. A prolific inventor, he holds 34 patents spanning memory systems, cache coherence, floating-point and integer execution units, and vector organization. Huffman has shared his insights as a speaker at the IEEE International SoC Conference and previously served as Chief Architect at Cadence Design Systems.
Jerry Ardizzone
Business Partner
Jerry Ardizzone is a seasoned sales executive with a proven track record in the semiconductor industry, having previously served as Executive VP of Worldwide Sales at ARM, where he secured multi-million-dollar contracts with top chipmakers. Now partnering with Artth to drive commercialization efforts, he brings decades of high-level sales expertise from leadership roles at Imagination Technologies, Codasip, and PIDA. His deep industry connections and deal-making experience position him as a key asset in bringing cutting-edge technologies to market.
