Server management gets its own card

For decades, server motherboards carried every critical subsystem on one large printed circuit board. That includes the baseboard management controller (BMC), a specialized processor that handles remote access, power control, sensor monitoring, firmware updates, system event logging, and error reporting. The BMC and its supporting circuitry typically sat alongside the host CPU, memory, and board logic. As high-speed interfaces push motherboard designs toward more expensive materials and additional layers, keeping management and security functions on that same board has become a cost and flexibility problem.

The Datacenter-ready Secure Control Module (DC-SCM) decouples server management and security from the host processor board. Cloudflare, in partnership with Lenovo, has now published a full DC-SCM design specification under the Open Compute Project (OCP) contribution database under the name Project Argus.

Why split the board?

Modern server memory and I/O buses are the main driver behind rising motherboard costs. DDR5 currently operates at 6400MT/s, with future speeds expected to reach 7200MT/s or 8800MT/s. PCIe Gen5 doubles the Gen4 signaling rate to 32 GT/s. These interfaces force motherboard designs onto ultra-low-loss dielectric materials with higher layer counts. The BMC and its surrounding circuits do not need that kind of performance, however. The ASPEED AST2600 BMC, for instance, still uses a PCIe Gen2 (5 GT/s) physical layer.

By moving management circuitry onto a separate, smaller module, the host processor module (HPM) printed circuit board can shrink, reducing the amount of expensive ultra-low-loss material required. The DC-SCM itself can be built on a smaller board with fewer layers and less costly low-loss dielectric. In Cloudflare's next-generation server design, the HPM board requires 14 or more layers while the DC-SCM routes comfortably on 8 to 10 layers.

Separation also creates flexibility. Management and security functions move to a common add-in card form factor, letting platform developers customize security or management for specific customers without spinning up multiple motherboard variants. A DC-SCM can carry over between server generations when management requirements are unchanged, and it can be swapped within a generation to upgrade management hardware across many platforms in a data center.

One firmware image, many platforms

Standardization of the interface between DC-SCM and HPM, known as the data center secure control interface (DC-SCI), enables electrical interoperability between different module and host designs. That interoperability opens the door to a unified firmware approach. Instead of maintaining a separate BMC image per server platform, a single OpenBMC image can serve multiple platforms. At boot, the device tree for the specific server is loaded automatically based on product information.

The public OpenBMC repository is the collaboration point for firmware development, and a single image simplifies firmware release management across an entire fleet, since one release propagates changes to every supported platform.

The Project Argus implementation

The DC-SCM 2.0 specification leaves certain implementation details open. Project Argus, Cloudflare's first DC-SCM fully compliant with that spec, documents the specific choices made for those items, covering form factor, mechanical locking, faceplate design, DC-SCI pinout, BMC and hardware root of trust (HWRoT) pinouts, and the minimum bootable device tree.

Project Argus is a horizontal External Form Factor (EFF) module built around the ASPEED AST2600 BMC System-on-Chip, which runs OpenBMC and provides the standard suite of remote management features. Security is handled by the ASPEED AST1060 as the HWRoT, providing firmware authentication, recovery, and update functions. A Lattice MachXO3D CPLD with secure and dual boot capability supports several IO interfaces, including LTPI, SGPIO, UART, and GPIOs.

Cloudflare and Lenovo have contributed the Project Argus design specification and reference design files to the OCP database, including:

  • SPI, I2C/I3C, UART, LTPI/SGPIO block diagrams
  • DC-SCM PCB stackup and board placements (TOP and BOTTOM layers)
  • DC-SCM schematic PDF file
  • DC-SCI pin definition PDF file
  • Power sequence PDF file
  • DC-SCM bill of materials spreadsheet
  • Mechanical drawings for the card assembly and interlock rail
  • Minimum bootable device tree requirements

The Project Argus DC-SCM 2.0 is slated for integration into Cloudflare's next-generation Gen 12 servers.