The Science Behind Dynex

Patent: Method and System for Large-Scale Computation of Quantum Algorithms; WIPO; International Application Number WO24231907; 24 Sep 2024; Abstract (EN): Current quantum computing technology, while promising, faces significant challenges that limit its scalability and practical application. These challenges primarily stem from the limited number of qubits available, the high susceptibility of qubits to errors, and the complexity of error correction protocols necessary to maintain coherence in quantum states. As quantum systems grow in size, the need for sophisticated error correction becomes increasingly critical, further complicating the development of reliable quantum computers. The invention described enables efficient, large-scale computation of quantum algorithms and circuits on traditional hardware, without sacrificing the fidelity or capabilities of quantum mechanics-based systems, thus bridging the gap between classical and quantum computing paradigms.

Patent: Quantum-Optimized Large Language Model for Non-Sequential Text Synthesis; WIPO; International Application Number PCT/IB2025/056653; 30 June 2025; Abstract (EN): A quantum diffusion large language model (QD-LLM), exemplified by Dynex DQLLM 1.1B, provides non-sequential text synthesis through diffusion-based text generation and quantum computing to solve the QUBO problem. The QUBO problem is solved using quantum computing or quantum-inspired algorithms, achieving 10x faster text generation and 10x lower computational costs than autoregressive models, and 5x efficiency gains over classical diffusion models like LLaDA. QD-LLM yields a HellaSwag score of 72.0 (+10.6 pp over TinyLlama v1.1), a TruthfulQA score of 54.5 (0-shot), and significant accuracy gains (e.g., +23.2 pp on BoolQ), outperforming ARM and LLaDA. The system is integrated into an API for scalable deployment in enterprise and cloud environments.

Patent: Method and System for Large-Scale Computation of Quantum Algorithms; USPTO; Application Number 63/948,358; 2025; (EN) The embodiments of this invention pertain to the domain of quantum computing, specifically addressing methods, apparatus, and systems for large-scale computation of quantum algorithms and quantum circuits. These embodiments leverage an innovative neuromorphic quantum computing architecture, designed to enhance computational efficiency and scalability in simulating and executing complex quantum processes.

Patent: Quantum-Optimized Large Language Model for Non-Sequential Text Synthesis; USPTO; Application Number 63/948,370; 2025; (EN) This invention pertains to artificial intelligence, specifically large language models (LLMs) for natural language processing (NLP). It introduces the Quantum Diffusion Large Language Model (QD-LLM), a novel framework that combines diffusion-based text generation with quantum computing to solve the Quadratic Unconstrained Binary Optimization (QUBO) problem for non-sequential text synthesis. By leveraging quantum computing to optimize token selection, QD-LLM achieves superior coherence, creativity, factual accuracy, and computational efficiency compared to both autoregressive LLMs and classical diffusion-based models.

Patent: Distributed Neuromorphic Supercomputing Platform (分布式神经形态超级计算平台), Publication Number CN116384458A, Publication Date 2023.07.04, China National Intellectual Property Administration; Abstract (EN): The present invention discloses a distributed neuromorphic supercomputing platform, which relates to the field of distributed computing technology; it includes a control unit, a group of interconnected storage units, logic gates and arithmetic functions, the storage unit can perform digital or analog operations under the control of the control unit, the storage unit is a memristor, the platform machine of the distributed neuromorphic supercomputing platform is implemented based on a GPU chip, the memristor adopts a self-organizing logic gate framework, and the self-organizing logic gate framework uses basic instantons to reach an equilibrium point. The present invention uses a GPU to achieve real-time performance close to that of a platform machine. It can be freely adapted to the problem to be calculated, and can also be interconnected and operated as part of a cluster. Using platform neuromorphic chips, problems that cannot be solved by classical or quantum methods can be solved, thereby improving the efficiency of use, simplifying the way data is transmitted, and liberating the performance of the machine.

Neuromorphic Computing for Computer Scientists: A complete guide to Neuromorphic Computing on the Dynex Neuromorphic Cloud Computing Platform

Dynex Developers; Amazon; 2024; ISBN-13: ‎979-8874282196

Applied Quantum Computing with Dynex: From Qubits to Unlocking the Future of Computing

Dynex Moonshots Foundation; Amazon; 2024; ISBN-13: ‎979-8874282196

计算机科学家的神经形态计算：在Dynex神经形态云计算平台上的神经形态计算完全指南

Dynex Developers; Amazon; 2024; ASIN: ‎B0CSBPR9WL

Advancements in Unsupervised Learning: Mode-Assisted Quantum Restricted Boltzmann Machines Leveraging Neuromorphic Computing on the Dynex Platform; Adam Neumann, Dynex Developers; International Journal of Bioinformatics & Intelligent Computing. 2024; Volume 3(1):91- 103, ISSN 2816-8089; Abstract: The integration of neuromorphic computing into the Dynex platform signifies a transformative step in computational technology, particularly in the realms of machine learning and optimization. This advanced platform leverages the unique attributes of neuromorphic dynamics, utilizing neuromorphic annealing - a technique divergent from conventional computing methods - to adeptly address intricate problems in discrete optimization, sampling, and machine learning. Our research concentrates on enhancing the training process of Restricted Boltzmann Machines (RBMs), a category of generative models traditionally challenged by the intricacy of computing their gradient. Our proposed methodology, termed “quantum mode training”, blends standard gradient updates with an off-gradient direction derived from RBM ground state samples. This approach significantly improves the training efficacy of RBMs, outperforming traditional gradient methods in terms of speed, stability, and minimized converged relative entropy (KL divergence). This study not only highlights the capabilities of the Dynex platform in progressing unsupervised learning techniques but also contributes substantially to the broader comprehension and utilization of neuromorphic computing in complex computational tasks.

HUBO & QUBO and Prime Factorization; Samer Rahmeh, Cali Technology Solutions, Dynex Developers; International Journal of Bioinformatics & Intelligent Computing. 2024; Volume 3(1):45-69, ISSN 2816-8089; Abstract: This document details the methodology and steps taken to convert Higher Order Unconstrained Binary Optimization (HUBO) models into Quadratic Unconstrained Binary Optimization (QUBO) models. The focus is primarily on prime factorization problems; a critical and computationally intensive task relevant in various domains including cryptography, optimization, and number theory. The conversion from Higher-Order Binary Optimization (HUBO) to Quadratic Unconstrained Binary Optimization (QUBO) models is crucial for harnessing the capabilities of advanced computing methodologies, particularly quantum computing and DYNEX neuromorphic computing. Quantum computing offers potential exponential speedups for specific problems through its intrinsic parallelism capabilities. Conversely, DYNEX neuromorphic computing enhances efficiency and accelerates the resolution of intricate, pattern-oriented tasks by simulating memristors in GPUs, employing a highly decentralized approach, via Blockchain technology. This transformation enables the exploitation of these cutting-edge computing paradigms to address complex optimization challenges effectively. Through detailed explanations, mathematical formulations, and algorithmic strategies, this document aims to provide a comprehensive guide to understanding and implementing the conversion process from HUBO to QUBO. It underscores the importance of such transformations in making prime factorization computationally feasible on both existing classical computers and emerging computing technologies.

Quantum-Driven Neuromorphic Computing for Million-Qubit-Scale Workloads; Adams Ivanov, Samer Rahmeh, Daniela Herrmann, Dynex Holding Establishment; Erick Giovani Sperandio Nascimento, University of Surrey; 145805582; Academia.edu; 2026; Abstract: We introduce Apollo, a 10,000-node p-qubit neuromorphic processor fabricated in 16 nm mixed-signal CMOS, operating fully at room temperature with a typical analog-core power envelope of ~0.5 W. The fundamental computing element, termed a p-qubit (probabilistic qubit), is a bistable stochastic unit whose continuous-time state fluctuations are driven by dedicated Integrated Quantum Entropy Units (IQEUs) injecting true, non-deterministic quantum-derived entropy. This elevates conventional p-bit architectures into the quantum regime, enabling state-transition rates of approximately 12.5ps per p-qubit at an energy cost of ≤10 fJ per transition. Apollo implements a high-degree Hyperion Δ256 interconnect topology, allowing near-native embedding of dense Ising and quadratic unconstrained binary optimization (QUBO) problems with substantially reduced minor-embedding overhead relative to existing annealing platforms. Through the Suzuki–Trotter isomorphism, the equilibrium statistics and annealing dynamics of the p-qubit network reproduce those of a transverse-field quantum annealer in one fewer spatial dimension, without requiring cryogenic cooling, long-lived coherence, or microwave control infrastructure. Beyond device-level validation, we demonstrate quantum-advantaged dynamics at scale by reproducing the three-dimensional spin-glass benchmark previously used to establish quantum advantage in superconducting quantum annealers. Across 300 disorder realizations, Apollo exhibits residual-energy scaling trajectories indistinguishable from those reported for cryogenic quantum annealing hardware and clearly distinct from simulated quantum annealing and classical simulated annealing, indicating access to the same quantum-critical dynamical universality class. A 350 nm release-candidate device (Apollo-RC1) experimentally validates the core p-qubit dynamics, thermodynamic sampling correctness, and continuous-time annealing behavior, while large-scale results establish Apollo as the first room-temperature, industrially scalable platform to demonstrate quantum-equivalent annealing performance on a canonical hard optimization benchmark. By unifying probabilistic computing, quantum-driven stochastic dynamics, and gate-compatible operation in a single architecture, Apollo opens new pathways for energy-based optimization, Bayesian inference, generative modeling, and hybrid classical–quantum workflows beyond the cryogenic era.

Benchmarking the Dynex Quantum Platform: A Comparative Evaluation of Quantum Algorithm Performance and Scaling; Adams Ivanov, Samer Rahmeh, Daniela Herrmann, Dynex Holding Establishment; 2026; Abstract: This paper presents an independent and rigorous evaluation of the Dynex quantum emulation platform, emphasising its performance across six benchmark quantum algorithms in comparison to leading quantum computing environments, including IBM Quantum, IonQ, and D-Wave. The benchmarks encompass n-Bit Adder Circuits, the Traveling Salesman Problem (TSP), Reverse Circuit Execution, Shor’s Algorithm, Protein Folding, and Reverse Hashing using Grover’s Algorithm. Each problem is formally introduced with its computational complexity class, highlighting the theoretical underpinnings and relevance of quantum approaches. Empirical results demonstrate that Dynex achieves problem sizes orders of magnitude larger than baselines, with runtime scaling that is sub-exponential—often polynomial or near-linear—contrasting the exponential constraints of gate-based and annealing systems. Mathematical analyses, including curve fitting and complexity derivations, confirm these scaling behaviours. All evaluations utilise replicable Python scripts on the Dynex SDK, ensuring methodological transparency. These findings substantiate Dynex's ODE-based emulation as a superior alternative for scalable quantum algorithm execution on classical hardware, with implications for advancing quantum computing research.

Harnessing Quantum Diffusion Extreme Weather Forecasting to Optimize WTI Trading with Machine Learning; Hok Chi Au, Daniela Herrmann, Stan Chen, Samer Rahmeh, Dynex Developers; 128316798 Academia.edu; 2025; Abstract: In the rapidly evolving domain of energy trading, integrating cutting-edge meteorological forecasting presents a transformative opportunity to refine commodity trading strategies. This paper details our collaboration between Dynex and Recycgo Go in developing an enhanced quantum weather prediction system that employs a diffusion model, achieving unprecedented forecast accuracy of up to 98% over a 14-day period. We provide a comprehensive background on the intersection of extreme weather events and WTI crude oil market dynamics, highlighting how highly accurate forecasts can inform risk mitigation, optimize decision-making, and capture opportunities driven by weather-induced price volatility. Furthermore, we discuss the design and implementation of a machine learning trading bot, capable of adapting its strategies in real time based on forecast data, thereby pushing the boundaries of algorithmic commodity trading.

Approaching Google Willow Chip’s Beyond-Classical Random Circuit Sampling Benchmarks Using Dynex; Adam Neumann, Dynex Developers; 126456331; Academia.edu; 2024; Abstract: Random Circuit Sampling (RCS) has emerged as a key benchmark for demonstrating quantum computational advantage. In this work, we implement RCS protocols on the Dynex platform, achieving results that approach Google’s beyond-classical benchmarks on their Willow chip. We demonstrate successful RCS execution on both 4 4 and 10 10 qubit grids, with circuit depths ranging from 5 to 200 cycles. Using patch-based cross-entropy benchmarking (XEB), we verify non-trivial fidelities across these configurations. Our complexity analysis shows that classically sim- ulating these circuits would require billions of years on current supercomputers, while our quantum implementation completes in minutes. Through careful validation and complexity estimation, we provide strong evidence that the Dynex platform has achieved quantum supremacy. The accessibility of our platform and public availability of our complete codebase enables independent verification of these results by the broader scientific community. Our work demonstrates that beyond-classical quantum computation is not limited to specialized hardware but can be achieved on more accessible platforms, marking an important step toward practical quantum computing applications.

Integration of LifeRacing Systems with Dynex n.quantum Platform; Samer Rahmeh, Head of Quantum Solutions Architecture, Dynex; Adam Neumann, Dynex Developers; 124541814; Academia.edu; 2024; Abstract: This paper presents an approach to optimizing gear shift strategies in racing ve- hicles by integrating LifeRacing Systems with the Dynex n.quantum platform. We propose a quantum-classical hybrid algorithm that leverages high-frequency teleme- try data and quantum annealing techniques to discover optimal gear shift patterns. The integration aims to enhance racing performance through more efficient and adaptive gear shifting strategies.

Computing a Quantum Volume of 2^119 Using the Dynex Neuromorphic Quantum Computing Platform; Samer Rahmeh, Head of Quantum Solutions Architecture, Dynex; Adam Neumann, Dynex Developers; 124276903; Academia.edu; 2024; Abstract: Quantum Volume (QV) is a holistic benchmark that measures the performance of quantum computers, accounting for both gate fidelity and circuit complexity. Achieving a high QV is essential for demonstrating quantum advantage over classical systems. In this paper, we present a detailed account of computing a Quantum Volume of 2^119 using the Dynex neuromorphic quantum computing platform. We describe the methodologies employed, the modifications made to standard QV testing protocols to accommodate the Dynex architecture, and the results obtained. Our work showcases the scalability and computational capabilities of the Dynex platform in handling large-scale quantum computations.

Quantum Frontiers on Dynex: Elevating Deep Restricted Boltzmann Machines with Quantum Mode-Assisted Training; Adam Neumann, Dynex Developers; 116660843, Academia.edu; 2024; Abstract: In this paper, we unveil an innovative methodology that marries quantum computing with the training of Deep Restricted Boltzmann Machines(Deep RBMs), marking a significant leap forward in deep learning technologies. Our approach, Quantum-Enhanced Mode-Assisted Training, builds upon the foundation laid by our previous work in single-layer RBMs, scaling this technique to the intricate architectures of Deep RBMs with unprecedented efficiency. By harnessing quantum states, our algorithm navigates the training landscape of Deep RBMs with a finesse that outstrips tradi- tional methods, capitalizing on the inherent parallelism and high-dimensional state representation of quantum computations. This quantum intervention not only preserves but notably enhances the benefits of mode-assisted training—yielding improved convergence rates, heightened stability, and an exponential reduction in the model’s parameter count without diminishing its representational capabilities. Remarkably, this quantum-enhanced approach to Deep RBMs drastically curtails the parameter space required, eclipsing the efficiency of both classical Deep RBMs and their quantum-trained counterparts using conventional techniques. This significant reduction in parameter de- pendency, achieved without sacrificing model fidelity, represents a groundbreaking advancement in modeling complex, high-dimensional data sets with Deep RBMs. Our findings promise to redefine the paradigms of unsupervised learning, especially within quantum computing frameworks, highlighting the superior capability of our method to not only advance RBM training but to revolutionize Deep RBM applications even further. This research is powered by Dynex, a cutting-edge quantum supercomputing platform, which enables us to leverage the full potential of quantum computing to its utmost advantage.

Framework for Solving Harrow-Hassidim-Lloyd Problems with Neuromorphic Computing using the Dynex Cloud Computing Platform; Samer Rahmeh, Cali Technology Solutions, Dynex Developers; 112871175; Academia.edu; 2023; Abstract: This document provides my steps and methodology on converting the Harrow-Hassidim-Lloyd (HHL) algorithm, typically used for solving linear systems on quantum computers, into a Quadratic Uncon- strained Binary Optimization (QUBO) model termed as ”QCFD” to be computed on DYNEX Neu- romorphic Network. This adaptation allows the use of classical and quantum-inspired solvers (a.k.a Simulated Annealing Sampler) and DYNEX Network users for finding solutions.