Fujitsu's PHOTON Architecture Boosts AI Performance 475x, Tackles Compute Bottlenecks
As large models continue to evolve rapidly, computing costs and processing efficiency remain key industry concerns. Fujitsu recently introduced a novel architecture called PHOTON (Top-down Network Parallel Hierarchical Computing), designed to overcome the performance limitations of traditional Transformer models in complex scenarios.
The Transformer architecture, now dominant in AI, handles long texts or high-concurrency multi-query tasks well, but often slows down due to frequent memory access for retrieving historical information, increasing GPU workload. Fujitsu's research team addressed this pain point by rethinking the PHOTON architecture's design.

PHOTON's core strength lies in its hierarchical processing approach. Instead of the token-level segmentation used by traditional Transformers, PHOTON introduces semantic layering, which cuts computational complexity while boosting parallel computing. Additionally, for multi-query tasks, the architecture streamlines decision-making: it needs only a single inference to reach a conclusion, using "majority voting" or "best choice" strategies.
Test results show that in smaller models with 600M, 900M, and 1.2B parameters, PHOTON delivers very high throughput and extremely low memory usage. In the 1.2B parameter model, its multi-query performance reaches 475 times that of mainstream Transformer architectures, significantly improving resource scheduling efficiency.
Because this architecture requires less KV Cache per iteration, the system can handle more iterations. This is a major performance boost for intelligent agent systems that manage many I/O processes. While some quality metrics see a slight trade-off, PHOTON's leap in computational efficiency makes it a promising technical solution for reducing AI operational costs.
Fujitsu is now actively promoting the architecture's application, aiming to provide lighter, more efficient underlying support for future intelligent scenarios through innovations in core algorithms.
Related article
U.S. Stocks Hit Historic Milestone as AI and Aerospace Giants Prepare for Trillion-Dollar Debut
Elon Musk, Sam Altman, and Dario Amodei, three titans of the technology sector, are advancing toward initial public offerings for their respective ventures. With SpaceX, OpenAI, and Anthropic—three industry behemoths nearing trillion-dollar valuation
Swedish AI Startup Lovable Eyes $13.2 Billion Valuation After Major Funding Round
As AI-driven coding tools gain traction, Swedish startup Lovable has secured a major funding round. The company aims to raise $3 billion, potentially boosting its valuation to $13.2 billion—double the $6.6 billion recorded last December. Menlo Ventur
Google Tests Remy AI Agent for Gemini as Focus Shifts to User Control
According to Business Insider, Google is testing Remy, a new AI personal agent for Gemini. This tool aims to execute tasks on behalf of users, streamlining both professional workflows and daily routines.Currently, Remy is undergoing testing in an int
Related Special Topic Recommendations
Comments (0)
0/500
As large models continue to evolve rapidly, computing costs and processing efficiency remain key industry concerns. Fujitsu recently introduced a novel architecture called PHOTON (Top-down Network Parallel Hierarchical Computing), designed to overcome the performance limitations of traditional Transformer models in complex scenarios.
The Transformer architecture, now dominant in AI, handles long texts or high-concurrency multi-query tasks well, but often slows down due to frequent memory access for retrieving historical information, increasing GPU workload. Fujitsu's research team addressed this pain point by rethinking the PHOTON architecture's design.

PHOTON's core strength lies in its hierarchical processing approach. Instead of the token-level segmentation used by traditional Transformers, PHOTON introduces semantic layering, which cuts computational complexity while boosting parallel computing. Additionally, for multi-query tasks, the architecture streamlines decision-making: it needs only a single inference to reach a conclusion, using "majority voting" or "best choice" strategies.
Test results show that in smaller models with 600M, 900M, and 1.2B parameters, PHOTON delivers very high throughput and extremely low memory usage. In the 1.2B parameter model, its multi-query performance reaches 475 times that of mainstream Transformer architectures, significantly improving resource scheduling efficiency.
Because this architecture requires less KV Cache per iteration, the system can handle more iterations. This is a major performance boost for intelligent agent systems that manage many I/O processes. While some quality metrics see a slight trade-off, PHOTON's leap in computational efficiency makes it a promising technical solution for reducing AI operational costs.
Fujitsu is now actively promoting the architecture's application, aiming to provide lighter, more efficient underlying support for future intelligent scenarios through innovations in core algorithms.
U.S. Stocks Hit Historic Milestone as AI and Aerospace Giants Prepare for Trillion-Dollar Debut
Elon Musk, Sam Altman, and Dario Amodei, three titans of the technology sector, are advancing toward initial public offerings for their respective ventures. With SpaceX, OpenAI, and Anthropic—three industry behemoths nearing trillion-dollar valuation
Swedish AI Startup Lovable Eyes $13.2 Billion Valuation After Major Funding Round
As AI-driven coding tools gain traction, Swedish startup Lovable has secured a major funding round. The company aims to raise $3 billion, potentially boosting its valuation to $13.2 billion—double the $6.6 billion recorded last December. Menlo Ventur





Home






