# Primemas

> Primemas is a Santa Clara fabless semiconductor company building Hublet, a reusable SoC foundation that packages the common plumbing of a custom chip - CPU cores, memory controllers, I/O, security, software and die-to-die links - so customers can attach their own accelerator rather than start over. Its Falcon-1 silicon now underpins CXL memory cards, switchless pools and rack-scale JBOM systems for AI, HPC and data-heavy computing.

- **Founded:** 2023
- **Headquarters:** Santa Clara, United States
- **Founders:** Il (Will) Park (Founder and CEO)
- **Products:** Hublet platform, Falcon-1 / PMH-1200, Kameleon, PMA14 and PMA16 CXL memory AICs, SLiM
- **Notable:** Delivered engineering samples and development boards for a CXL 3.0 memory-controller SoC to strategic customers in 2025., Turned the Hublet concept into named Falcon-1 silicon and a product ladder spanning direct-attached memory, switchless pooling, JBOM and near-memory computing., Raised a $72 million Series B in January 2026.

## Products & services

- **Hublet platform** — A reusable hub-SoC architecture integrating CPU, memory controllers, I/O, security, software and high-speed die-to-die interfaces for attaching customer chiplets.
- **Falcon-1 / PMH-1200** — Primemas's chiplet-based platform SoC and CXL controller silicon, designed as the common foundation for memory systems, AI accelerators and server products.
- **Kameleon** — An FPGA companion chiplet for prototyping, field-adaptable features and custom acceleration; the platform incorporates Achronix Speedcore eFPGA technology.
- **PMA14 and PMA16 CXL memory AICs** — High-capacity add-in cards for direct-attached memory expansion; PMA16 is presented as a 16-DIMM card supporting up to 8TB with 512GB RDIMMs.
- **SLiM** — A switchless pooled-memory system intended to share terabyte-scale DDR5 capacity without a separate external CXL switch.
- **JBOM** — A rack-scale 'Just a Bunch of Memory' appliance that aggregates CXL memory for AI, HPC, database and analytics workloads, from tens of terabytes toward petabyte-scale configurations.
- **Physical AI acceleration platforms** — Mobile vision and multimodal edge-AI reference systems combining camera interfaces, NPU processing and high-capacity memory.
- **Prototyping, emulation and bespoke hardware** — Contact-led B2B platforms built from Falcon Hublets and Kameleon FPGA chiplets in custom-board, add-in-card and server-chassis formats.

## Achievements

- Delivered engineering samples and development boards for a CXL 3.0 memory-controller SoC to strategic customers in 2025.
- Turned the Hublet concept into named Falcon-1 silicon and a product ladder spanning direct-attached memory, switchless pooling, JBOM and near-memory computing.
- Raised a $72 million Series B in January 2026.
- Joined Micron and PNNL on the planned 100TB-class Abaco shared-memory deployment for scientific AI and HPC.
- Expanded a 2024 Samsung memorandum into a three-way 2026 CXL memory-system development effort with Samsung and ETRI.
- Demonstrated the complete CXL memory stack at FMS 2026 and showed PMA16 at the AI Infra Summit.

## Latest updates

- **2026-09** — Primemas showed PMA16 at AI Infra Summit: a 16-DIMM CXL Type-3 card listed at up to 8TB with 512GB RDIMMs, using dual CXL 3.2 links.
- **2026-08** — At FMS 2026, Primemas presented PMA14/PMA16 cards, SLiM switchless pooling, JBOM and the Abaco rack roadmap with Micron, PNNL and Liqid.
- **2026-04** — Samsung, ETRI and Primemas announced joint development of an ultra-high-capacity CXL memory-centric computing system.
- **2026-01** — Primemas closed a $72 million Series B round.
- **2025-06** — The company announced customer availability of its Falcon CXL 3.0 controller samples and collaboration with Micron's validation program.

## Links

- Website: https://www.primemas.com/
- LinkedIn: https://www.linkedin.com/company/primemas

---

Profile page: https://yespress.io/primemas
Published by YesPress — https://yespress.io
Last updated: 2026-09-26