CARLSBAD, Calif.–MaxLinear, Inc., a leading provider of high-speed interconnect ICs enabling data center, metro, and wireless transport networks, today announced the development of its fourth generation 200G/lane PAM4 SERDES (Serializer/Deserializer) and DSP family, Rushmore. This new Samsung Foundry leading-edge-CMOS-based family delivers a significant leap in performance, power efficiency and density, enabling data center operators and cloud providers to address rapidly increasing data traffic demand, the transition to higher speed networks, and the critical power efficiency needs of AI/ML networks.
“This collaboration signifies our commitment to delivering cutting-edge solutions that empower our customers to build next-generation data center and high-performance computing applications.”
The Rushmore family encompasses a diverse array of variants tailored to meet the evolving needs of modern data centers and telecommunications networks. This includes support for fully re-timed optical modules, half re-timed optics, and active electrical cables (AECs). Moreover, MaxLinear is introducing both 8x200G-to-8x200G re-timers and 8x100G-to-4x200G gearbox variants, providing customers with a comprehensive suite of solutions to address their specific requirements.
“The AI/ML revolution has accelerated the pace of innovation across the industry, and that is also the case at MaxLinear,” said Drew Guckenberger, Vice President of High Speed Interconnect at MaxLinear. “Starting from our highly successful Keystone 5nm 100G/lane DSP family, the amount of additional innovation that has gone into the new Rushmore 200G/lane DSP family is incredible. The impacts on power and performance our partners will see are expected to be significant for the industry as a whole as we scale into the new AI/ML world.”
Building on the technical and commercial success of the Keystone family of 5nm 100G/lane DSPs, Rushmore doubles the data rates and significantly reduces latency for Big Data and AI/ML applications, while simultaneously adding significant additional functionality and performance extensions. The Rushmore family of DSP has been developed in conjunction with homegrown co-optimized SERDES to achieve overall best-in-class power and performance.
“We are excited to collaborate closely with MaxLinear on this industry-leading achievement. Their proven expertise in high-speed interconnect solutions, combined with their strong development approach, made them an ideal partner for developing this next generation PAM4 SERDES IP and DSP using our most advanced manufacturing CMOS process technology,” said Marco Chisari, Executive Vice President and Head of US Foundry Business, Samsung Electronics. “This collaboration signifies our commitment to delivering cutting-edge solutions that empower our customers to build next-generation data center and high-performance computing applications.”
AI and ML have emerged as key drivers for the growth of the global high-performance computing market, which is directly linked to the need for faster data processing and transfer for AI workloads. Training AI models can be computationally expensive and time-consuming. Higher data rates, enabled by technologies like 200G/lane SERDES and DSPs, allow for faster data movement between processing units and memory, accelerating the training process. Similarly, faster inference speeds are crucial for real-time applications like autonomous vehicles or image recognition, requiring efficient data transfer through high-bandwidth connections.
“Shipments of PAM4 DSPs will grow at a CAGR (Compound Annual Growth Rate) of 50% between 2023 and 2027, primarily driven by AI/ML applications,” said Vlad Kozlov at LightCounting. “MaxLinear, with its strong track record in high-speed interconnect solutions, is well-positioned to capitalize on this trend by delivering innovative 200G/lane PAM4 SERDES and DSP solutions that address the specific needs of the AI/ML market, such as high performance, low power consumption, and low latency.”
MaxLinear will be exhibiting at the Optical Fiber Communication (OFC) conference in San Diego, California, March 26-28, 2024 at booth 4501.
