Panmnesia Unveils Link Solution Product Lineup for AI Infrastructure

Fabless startup Panmnesia, Inc. unveils a comprehensive lineup of Link Solution products designed for AI infrastructure, reinforcing its technology brand identity.

Full-Stack Link Solution product lineup

The company is a startup dedicated to developing interconnect technologies for next-gen AI infrastructure. In particular, Panmnesia aims to improve AI infrastructure through the following approaches:

• Enabling composable architecture, allowing addition, removal, and reconfiguration of diverse devices such as GPUs, AI accelerators, and memory modules, to meet the evolving needs of AI applications.  
• Efficiently accelerating large-scale AI workloads by minimizing inter-device communication overhead.

Panmnesia’s Vision: Building better AI infrastructure by realizing composable
architecture and minimizing communication overhead

Link Solution Lineup for AI Infrastructure
On June 25, the company officially introduced its Link Solution product lineup, reinforcing its brand identity as a company specializing in AI infrastructure. This lineup encompasses the entire technology stack required for AI infrastructure design, including hardware, silicon IPs, network topology, and software. By delivering cross-stack optimization, Panmnesia’s Link Solution significantly reduces communication overhead and enhances overall system performance.

Another key feature of the Link Solution is its support for a wide range of connected devices and diverse interconnect methods, such as various network topologies and protocols. This enables users to flexibly configure and interconnect resources, adapting to the dynamic requirements of rapidly evolving AI workloads. To support this capability, the company is participating in various consortia as an official member, such as Compute Express Link (CXL) Consortium, UALink Consoritum, the Open Compute Project (OCP), and PCI-SIG.

List of consortia that Panmnesia currently participates in

Link Solution consists of following product categories:

Hardware Products
The hardware lineup includes two core elements essential for large-scale AI infrastructure connectivity: 1). Switch Chips (System-on-Chip, SoC) and 2) Retimer Chips.

The switch chip, one of Panmnesia’s flagship products, expands infrastructure from a device count perspective by enabling large-scale device connectivity and managing data transmission among connected components. The retimer chip, on the other hand, addresses physical distance challenges by restoring signal strength lost during long-distance communication. Both chips incorporate Panmnesia’s proprietary low-latency design IPs, allowing customers to reduce communication overhead significantly by building their core interconnect pathways with these products.

Silicon IP Products – Meanwhile, the company’s Link Solution is not limited to the core interconnect pathways. The company also offers silicon IPs (semiconductor IPs) targeting a wde range of system devices that make up AI infrastructure — including CPUs, GPUs, NPUs, memory, and storage.
Device manufacturers can use these IPs as design building blocks to enable their system devices to support specific interconnect protocols. Notably, each IP design is optimized for its target device, enabling customers to develop faster and more efficient hardware.

Custom Design Solutions
Finally, the firm provides custom design services across multiple layers — from silicon-level to network and software architecture — to help optimize AI infrastructure based on each client’s specific requirements.

These include:

• PanEP: Custom silicon services that produce AI accelerator or memory chips tailored to client needs
• PanFabric: Network architecture design services that minimize communication overhead considering characteristics of AI applications
• Total AI Solution: Comprehensive design service covering everything from semiconductor chips to systems and software.

With these services, customers can achieve AI infrastructure performance levels that are difficult to realize with off-the-shelf products alone.

Product categories of Link Solution: Hardware, silicon IP, and custom design solution

Panmnesia’s Ongoing Efforts to Advance AI Infrastructure
In addition to developing these products, the company is also pursuing a range of initiatives to further advance the state of AI infrastructure. Earlier this year, the company launched a large-scale R&D project focused on AI data centers and is currently developing foundational technologies — including chiplets, manycore acceleration, and in-memory processing.

The company is also actively sharing its technical insights and long-term vision through a range of technical documents and demonstration videos. These materials cover topics such as GPU memory expansion via interconnect technologies and case studies on accelerating AI applications using interconnect technologies.

A Panmnesia spokesperson commented, “Our efforts to build a better AI infrastructure will continue to accelerate and expand,” and added, “We are actively hiring to further accelerate the development of our Link Solution business.”

Additional information about Panmnesia’s vision for AI infrastructure and details about its Link Solution product lineup can be found on the company’s newly updated website.