{"slug": "amd-expands-psiquantum-partnership-across-cpus-gpus-and-software", "title": "AMD expands PsiQuantum partnership across CPUs, GPUs, and software", "summary": "AMD plans to make an undisclosed equity investment in PsiQuantum and expand its partnership with the quantum computing company beyond Versal FPGAs into AMD EPYC processors, Instinct accelerators, and the Radeon Open Compute (ROCm) software stack, the two companies said. The collaboration covers co-designed infrastructure for fault-tolerant quantum computers, hybrid quantum-HPC-AI applications, and PRISM, a transformer-based machine-learning decoder that a PsiQuantum-AMD preprint reports matches integer-programming decoders on logical error rates for surface, color, and quantum LDPC codes while requiring substantially less processing time, including an implementation on an AMD VP1902 FPGA. \"Leadership in classical computing is essential to building systems at scale,\" AMD CTO Mark Papermaster said.", "body_md": "AMD plans to invest in PsiQuantum as the companies expand their collaboration from programmable control hardware into the CPUs, GPUs, and software needed to support large-scale quantum computing.\n\nThe expanded partnership will combine AMD EPYC processors, Instinct accelerators, and the Radeon Open Compute (ROCm) software stack with PsiQuantum’s silicon photonics platform. It builds on existing work using AMD Versal field-programmable gate arrays (FPGAs).\n\nThe companies will co-design infrastructure for fault-tolerant quantum computers and develop hybrid applications combining quantum computing, high-performance computing (HPC), and AI. They did not disclose the size or terms of AMD’s planned equity investment.\n\n“Leadership in classical computing is essential to building systems at scale,” AMD CTO Mark Papermaster said.\n\n## Error correction needs fast classical compute\n\nOne strand of the collaboration concerns PRISM, a machine-learning decoder optimized for programmable hardware accelerators.\n\nA [research preprint from PsiQuantum and AMD researchers](https://arxiv.org/abs/2611.13630) describes the transformer-based decoder’s ability to interpret errors in photonic quantum computing architectures. Photon loss and probabilistic operations create specific demands: the decoder must remain accurate while handling missing quantum information at high rates.\n\nThe researchers benchmarked PRISM on protocols using surface codes, color codes, and quantum low-density parity-check codes. Across those benchmarks, they report logical error rates matching integer-programming-based decoders while requiring substantially less processing time.\n\nThe work also explores implementation on an AMD VP1902 FPGA, addressing the latency and throughput requirements of real-time decoding. Those requirements make classical processing part of the quantum computer’s operating infrastructure, rather than simply a resource used before or after a calculation.\n\n## Photonics reaches beyond quantum systems\n\nPsiQuantum is also pursuing applications for its photonics platform in conventional AI infrastructure.\n\nThe company recently [unveiled an optical circuit switch demonstrator](https://www.sdxcentral.com/news/psiquantum-brings-quantum-photonics-to-ai-network-switching/) designed to route traffic between eight inputs and eight outputs without optical amplification or moving parts. Company testing reported typical insertion loss of 1.2 dB and reconfiguration in less than 3 milliseconds.\n\nPsiQuantum is sharing samples with prospective customers ahead of planned commercial products. The device applies technology developed for quantum computing to connectivity between conventional computing resources, where optical switching could reduce conversion overheads.\n\nThe AMD collaboration also includes computational chemistry work already running on classical hardware. The companies report initial results from simulations of covalent inhibitors, molecules relevant to oncology drug discovery, and plan further work on pharmaceutical applications.\n\nThey will also explore expanding their collaboration at the Illinois Quantum and Microelectronics Park in Chicago, where PsiQuantum is an anchor tenant.", "url": "https://wpnews.pro/news/amd-expands-psiquantum-partnership-across-cpus-gpus-and-software", "canonical_source": "https://www.sdxcentral.com/news/amd-expands-psiquantum-partnership-across-cpus-gpus-and-software/", "published_at": "2026-11-18 06:10:40+00:00", "updated_at": "2026-11-18 06:14:34.094583+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "ai-infrastructure", "ai-chips"], "entities": ["AMD", "PsiQuantum", "Mark Papermaster", "EPYC", "Instinct", "ROCm", "PRISM", "Illinois Quantum and Microelectronics Park"], "also_reported_by": [], "takeaway": "PRISM, a transformer-based decoder for photonic quantum error correction, matched integer-programming decoders' logical error rates on surface, color, and quantum LDPC codes while needing substantially less processing time, with an implementation on an AMD VP1902 FPGA.", "recommendation": {"advice": "For real-time decoding workloads, evaluate transformer-based decoders like PRISM on programmable accelerators such as the AMD VP1902 FPGA; the preprint reports integer-programming-level logical error rates at substantially lower processing time.", "playbook": "inference-serving", "playbook_url": "https://wpnews.pro/playbooks/inference-serving"}, "alternates": {"html": "https://wpnews.pro/news/amd-expands-psiquantum-partnership-across-cpus-gpus-and-software", "markdown": "https://wpnews.pro/news/amd-expands-psiquantum-partnership-across-cpus-gpus-and-software.md", "text": "https://wpnews.pro/news/amd-expands-psiquantum-partnership-across-cpus-gpus-and-software.txt", "jsonld": "https://wpnews.pro/news/amd-expands-psiquantum-partnership-across-cpus-gpus-and-software.jsonld"}}