[Blog] A Layered Approach to AI Datacenter Control Plane Architecture

The expansion of artificial intelligence (AI) solutions is driving unprecedented change in datacenter infrastructure. These changes are not confined to raw compute power, either: platform complexity, power density, thermal management needs, and security demands all continue to evolve to support high volume, complex AI workloads,

In a recent Security Seminar, experts from Lattice, ASPEED, and AMI discussed growing infrastructure demands, their impact on datacenter design, and why strategic, layered platform control plane models are integral to scalable, resilient, and secure AI ecosystems.

The Platform Control Plane is Critical to AI Infrastructure
As AI infrastructure expands to meet growing demands, success depends on more than simply boosting compute performance. The core pressures on datacenter infrastructure can be divided into three categories:

1. Platform complexity. Modular, disaggregated AI architectures are creating more heterogeneous environments that require coordinated management across varied compute, memory, accelerators, and other components/subsystems.
2. Security. Security is now a foundational, device-level requirement. Hardware Root of Trust (HRoT), attestation, secure updates, and crypto-agility are no longer optional features; they are expected to be part of the platform architecture itself.
3. Power and thermal demands. Higher rack power densities and the rise of liquid cooling introduce new operational challenges into datacenter management, making uptime, fault detection, and coordinated control more important than ever.

These shifts fundamentally change how organizations evaluate their datacenter infrastructure. They’ve triggered the strategic reevaluation and fortification of AI systems’ platform control planes, the always-on layers responsible for system management, security, fault recovery, and uptime.

Control plane capabilities are now an early design consideration rather than an afterthought, with security architects, operations leaders, site reliability engineers, and platform teams collaborating to design the most effective models. When executed strategically, control planes can help support more secure, efficient AI infrastructure at scale.

Building a Layered Control Plane Architecture
To address this growing need and avoid reliance on a single limited component, Lattice, ASPEED, and AMI are working together toward a unified control plane architecture. Each layer of the model is optimized for a specific function, while still working collaboratively through open, standards-based interfaces.

This architectural vision brings together three key layers:

1. Deterministic Platform Control with Lattice FPGAs. As the foundation of this model, Lattice Field Programmable Gate Arrays (FPGAs) provide a first-on, last-off control layer that keeps critical platform functions available throughout the system lifecycle. This layer is responsible for power sequencing, board-level coordination, protocol bridging, and real-time fault response. It also enables teams to adapt to changing hardware requirements through its inherent reprogrammability.
2. System Management and Security with ASPEED BMC/SMC. ASPEED’s Baseboard Management Controller (BMC) and, in some configurations, its Satellite Management Controller (SMC), form the next layer of the model, providing out-of-band management, telemetry collection, and system-level control capabilities. This layer helps extend manageability beyond individual servers to support multi-node and rack-scale environments while delivering integrated security features like HRoT, secure boot, attestation, and secure update mechanisms. A concrete example of this collaboration is ASPEED’s AST1840, a recently announced device that integrates ASPEED's SMC with Lattice FPGA in a single package — combining tight board-level management with deterministic real-time control.
3. Firmware and Orchestration with AMI Solution. AMI solutions form the final layer, aggregating infrastructure data into a unified management view to enable fleet-level visibility, policy management, and orchestration integration through industry-standard interfaces, like DMTF’s Redfish^® and PLDM.

Combined, these three layers create a cohesive control plane that simplifies management across increasingly complex AI datacenter infrastructure.

What This Architecture Delivers for AI Infrastructure
This collaborative vision is rooted in years of domain expertise across three industry leaders. Lattice, ASPEED, and AMI are working closely with the goal of providing customers with reliable, interoperable control planes to help accelerate deployment while reducing integration risk.

The benefits of this approach are wide-ranging, including:

• Scalability. The architecture is designed to support increasingly modular and disaggregated architectures, while enabling coordinated management across compute, networking, storage, power, and cooling components.
• Reliability and uptime. FPGA-based control enables real-time response to power, thermal, and cooling needs, while redundant control paths help ensure that critical safety functions remain operational.
• Operational efficiency. By unifying management and orchestration, the model brings server, rack, and pod details into a single view. This simplifies tasks like monitoring, policy enforcement, and other continuous operations.
• Future-ready flexibility. Reprogrammable hardware and standards-based interfaces ensure platforms can adapt to evolving needs and requirements without extensive and expensive redesigns.
• Enhanced security. HRoTs, secure boot, attestation, secure updates, and crypto-agility could help create a consistent security posture across AI infrastructure. Lattice FPGAs can serve as a Platform Root of Trust (PRoT) or a 'symbiont root of trust', complementing ASICs like management controllers (MCUs), CPUs, or GPUs with hardware-based crypto-agility, PQC (post-quantum cryptography) readiness, and attestation support without requiring a costly redesign or re-tapeout. This is particularly relevant as PQC readiness is increasingly appearing in procurement requirements across geographies.

Together, these capabilities reflect Lattice, AMI, and ASPEED’s shared vision to help operators reduce complexity, maximize uptime, and build greater foundational trust in their datacenter infrastructure.

Building the Foundation for Next-Gen Datacenters
The platform control plane has become a critical, strategic differentiator for datacenter operators. By bringing together the deterministic secure-control of Lattice FPGAs, ASPEED’s management and security components, and AMI’s orchestration software, organizations can build AI systems that are scalable, resilient, and secure.

To learn more about reducing complexity and accelerating AI infrastructure deployment, watch the full Security Seminar here. To dive deeper into this collaborative control plane architecture, visit our website and contact our team today.