Microchip Unveils Plug-In Timing Module for AI-Burdened Data Centers

Microchip recently announced a new plug-in timing module family, the MD-990-0011-B. The timing module enables seamless time synchronization for a wide variety of distributed computing systems operating under critical conditions.

Microchip MD-990-0011-B M.2 timing module.

Microchip collaborated with Intel on the M.2 form-factor timing module for integration into specific Xeon 6 system-on-chip (SoC) server platforms. It uses Intel’s virtualized radio access network (vRAN) technology to synchronize timing across large, distributed networks.

The Challenges of Time Synchronization

Timing in the distributed computational world of AI, 5G communications, and modern data centers is far more complex than sending a crystal-controlled square wave into a processor chip. AI data centers make multiple processors—up to millions of cores—operate as a single extended processing unit. Non-AI data centers process transactions and send communications that require precise time synchronization with other transactions worldwide. This type of distributed computing requires stability and accuracy within the computing box, as well as synchronization with systems throughout the data center and across the globe.

The MD-990-0011-B timing module integrates easily with Intel's Xeon 6 SoC server platforms. 

The theoretical minimum transmission speed from one side of the Earth to the other is around 70 milliseconds. That amount of latency does not impede voice communications but can significantly impact time-stamp-critical communications. Within data centers, AI parallel computation must be synchronized within a few nanoseconds.

The MD-990-0011 relies on a variety of timing protocols, including global navigation satellite systems (GNSS), synchronous Ethernet (SyncE), and precision time protocol (PTP). GNSS serves as the grandmaster clock for modern timing systems. PTP (IEEE 1588) manages timing at the packet level, while SyncE operates at the physical layer. The combination, along with holdover in the module, enables consistent high-performance synchronization.

Features Up to the Task

The MD-990-0011-B's primary clock synchronization (datasheet linked) comes from recovered GNSS 1-Pulse-Per-Second (PPS) signals. Recovered GNSS 1PPS is a clock signal with its rising edge aligned with the start of the UTC standard second, the coordinated universal time. GNSS 1PPS is typically synchronized to within less than 100 nanoseconds.

Microchip has developed two variants of the module. The MD-990-0011-BC01 delivers eight hours of holdover (timekeeping during network disruptions or GNSS outages), while the -BA01 has four hours of holdover. Microchip delivers this performance by combining three Microchip subcomponents: the ZL80132B synchronous Ethernet (SyncE) synthesizer, two OX-22x oven-controlled crystal oscillators (OCXOs), and an MCP9808 temperature sensor.

MD-990-0011-B block diagram. 

The ZL80132B SyncE has dual-clock PPLs, three digital PPLs (DPLLs), and five low-jitter synthesizers. The SyncE manages synchronization and holdover accuracy based on clock signals from the pair of OCXOs and the temperature sensor. Holdover accuracy is within +/- 1.5 microseconds over the specified four or eight hours.

The MD-990-0011-B module utilizes the 30 mm x 70 mm M.2 plug-in module form factor and is designed around the Intel Xeon 6 system-on-chip (SoC) processor. 

Typical use case in Intel Xeon 6 server. (Find enlarged image on page 6 of the datasheet.) 

Microchip designed the module for easy deployment and easy upgrade within compatible systems. The M.2 form factor allows for easy upgrade within motherboards that relied on the prior -A generation and for future generations of timers.

Product Availability

Both MD-990-0011-B variants, -BA01 and -BC01, are available in production quantities directly from Microchip and through its worldwide distribution network. The key components are also available from Microchip in discrete form.

All images used courtesy of Microchip.