DreamBig Semiconductor develops high-performance accelerator platforms using Chiplet Hub with 3D high-bandwidth memory (HBM) technology. The company created the open "MARS Platform," an open chiplet platform that enables customers to develop scalable and composable semiconductor solutions for applications such as LLMs, GenAI, data centers, Edge computing, and automotive sectors.
Launched in January 2024, the MARS Platform's Chiplet Hub features a memory-first architecture, providing direct access to SRAM, DRAM, CXL, and SSD memory tiers from all compute, accelerator, and networking chiplets. The platform solves key challenges of scaling up compute and scaling out networking for AI servers and accelerators.
DreamBig's Chiplet Hub supports universal architecture for CPU, AI, accelerators, I/O, networking, and memory chiplets within a unified platform. It incorporates hardware acceleration technologies like fully associative cache/memory management, direct memory access (DMA), and RDMA for efficient data transfer. The platform also features virtual PCIe/CXL switches for optimized resource allocation and algorithmic TCAM hardware acceleration for match/action processing when scaled out to the cloud.
Key competitors: None comparable (DreamBig Semiconductor is the only firm we have identified that specializes in providing a high-performance accelerator platform to optimize and scale AI hardware.)
By using this site, you agree to allow SPEEDA Edge and our partners to use cookies for analytics and personalization. Visit our privacy policy for more information about our data collection practices.