AMD’s Innovative D2D Interconnect Technology Set to Revolutionize Zen 6 Processors
Summary:
- AMD is transitioning to a new D2D interconnect technology for Zen 6 processors, moving away from traditional SERDES.
- The new "Sea-of-Wires" design offers significant power consumption reductions and improved latency.
- Innovations tested on the Strix Halo APU suggest promising advancements in energy efficiency and performance.
In an exciting development for technology enthusiasts and industry professionals alike, AMD is set to unveil its next-generation D2D interconnect technology, aimed at enhancing the performance of its Zen 6 processors. This innovative shift, discovered by prominent YouTuber @High Yield, will replace the existing SERDES architecture and has already demonstrated substantial power optimization and latency improvements during testing on the Strix Halo APU.
Transitioning from SERDES to "Sea-of-Wires"
Since the introduction of the Zen 2 architecture, AMD has relied on SERDES PHY technology for high-speed interconnects between chiplet core components. This process involves a serializer converting parallel data into a serial bit stream to transmit across packages to the I/O/SoC chip, followed by deserialization. However, over time, this method has revealed several shortcomings:
-
Energy Consumption Overhead: The serialization and deserialization processes introduce additional requirements for clock recovery, equalization, and encoding. These steps lead to unnecessary energy consumption.
- Communication Latency: The conversion process adds delays, which negatively impacts the efficiency of increasingly complex chip communications.
While this design served well under traditional processor architectures, the integration of new modules, such as Neural Processing Units (NPUs), necessitates lower latency and higher bandwidth interconnections between chips.
Strix Halo Explores New Ground
The Strix Halo APU serves as a testing ground for AMD’s groundbreaking interconnection method, utilizing TSMC’s InFO-oS (Substrate-based Fanout Integrated Package) and Redistribution Layer (RDL) technology. This advanced approach features:
-
Parallel Wire Configuration: Multiple small parallel wires are arranged on the interposer between the chip and substrate, enabling broad parallel port communication.
-
Elimination of Traditional SERDES: The conventional SERDES module is replaced with a rectangular mini pad array, reflecting typical fan-out structural characteristics.
- Reduction in Processing Steps: The new design eliminates the need for serialization and reverse serialization. This change significantly lowers power consumption and latency while enhancing bandwidth by increasing the number of available ports.
Challenges Ahead
Despite the promising advantages of the "Sea-of-Wires" technology, it also introduces new complexities in design. The industry observes with interest, anticipating that the innovative interconnect solutions demonstrated in the Strix Halo APU will pave the way for enhanced energy efficiency and performance balance in the upcoming Zen 6 processors.
AMD’s transition to this groundbreaking technology signifies a strategic shift toward meeting the demands of modern applications and chip interconnectedness. As the company progresses toward the launch of these processors, the tech community eagerly awaits the performance benchmarks and real-world applications of the new D2D interconnect technology.
Looking Ahead
With the prospect of improved efficiency and reduced latency, AMD’s D2D interconnect technology positions the company for competitive advantage in the semiconductor market. Stakeholders and consumers alike are keen to see how these advancements translate into tangible benefits for future computing devices. Stay tuned for further updates as AMD continues its journey toward integrating this revolutionary design into its processor lineup.
By keeping an eye on these developments, consumers can better understand how AMD’s innovations will enhance their computing experiences and enable new possibilities in technology.
