05/09/2025, 05:27 AM UTC
韩国CXL专家Panmesia获3000万美元融资，以重新定义AI基础设施架构Panmesia gets $30m to redefine AI infrastructure architecture
➀ Panmesia获3000万美元融资，将开发基于芯粒（chiplet）的模块化AI加速器，针对大语言模型等AI服务优化资源利用率；
➁ 采用可扩展的芯粒架构和内存计算技术，通过CXL协议实现计算与存储资源动态调配，显著降低数据传输能耗与成本；
➂ 集成众核架构与向量处理器芯粒，利用先进制程满足AI算力需求，提供灵活配置方案以适配不同工作任务。
➀ Panmesia secures $30 million to develop modular AI accelerators using chiplet technology, targeting LLMs and large-scale AI workloads with optimized resource utilization;
➁ Its flexible chiplet architecture integrates in-memory processing and CXL to reduce data movement, power consumption, and infrastructure costs while enabling scalability;
➂ The accelerator combines manycore and vector processor chiplets, leveraging advanced nodes to deliver high parallelism and tailored performance for diverse AI tasks.
---
本文由大语言模型（LLM）生成，旨在为读者提供半导体新闻内容的知识扩展（Beta）。
