This blog is co-authored by Srini Narayana, VP, Boot Firmware Technology Group, AMI and Marc Meunier, Director Hardware Ecosystem, Infrastructure, Arm.
The shift from monolithic SoC designs to chiplet-based architecture isn’t just a packaging innovation. It’s a fundamental rethinking of how custom silicon is designed, manufactured, and deployed. This transition is driven by the growing impracticality of scaling large monolithic dies at advanced nodes.
As die sizes increase, so do the costs, yield risks, and integration complexity. Chiplets offer a practical solution, breaking the system into smaller, functionally specialized components that can be manufactured on different process nodes and stitched together using high-speed interconnects.
But this modularity introduces a new class of system-level challenges, especially in terms of firmware. In traditional monolithic designs, firmware typically operates within a tightly coupled environment, with well-defined assumptions about hardware topology, boot order, and resource management. Chiplet-based systems disrupt those assumptions.
Now, firmware must manage a distributed set of silicon components, sometimes with each element designed by a different team, built on different nodes, with critical interface requirements, and then integrated together very late in the development cycle.
From a firmware perspective, chiplet architectures are not just modular building blocks — they must be dynamically orchestrated to achieve ultimate performance. Each chiplet may have its own power domain, boot sequence, security enclave, and thermal profile. Firmware must then be able to orchestrate these elements cohesively, ensuring that the system boots reliably, communicates securely, and operates efficiently across the heterogeneous system environment.
This is the point where integrated modular firmware platforms become indispensable. Rather than treating firmware as a monolithic afterthought, effective CSS-based chiplet system firmware must be able to incorporate:
Under this new paradigm, these capabilities aren’t just conveniences; they are necessities. Without them, firmware development becomes a bottleneck, risking project delays, redesigns, re-prototyping, and potential catastrophic integration failures.
As chiplet architectures become the norm for custom silicon in numerous applications starting with HPC, cloud, and edge domains – and soon likely spreading into telecom, automotive, industrial and even PCs - the role of firmware is shifting from the periphery of the design space into a very central function. No longer is it just about booting the system; firmware is now enabling the modular systems to be brought together in the first place. Firmware must abstract complexity, enforce security, and provide a stable foundation for innovation across the entire custom silicon design.
One of the most practical advantages of modular firmware is its ability to absorb and integrate third-party IP from across the silicon ecosystem. Whether it's PCIe, CXL, memory controllers, or telemetry modules, these components often come from different vendors, each with its own quirks and requirements. In a chiplet-based system, firmware must act as the glue that binds these disparate elements into a coherent platform.
This is where early validation becomes critical. By “left-shifting” firmware testing into simulation and emulation environments - well before initial tape-out - teams can catch integration issues early, before they become expensive silicon bugs. Working closely with tool vendors, firmware teams can deliver pre-validated stacks that reduce the risk of post-silicon surprises and help avoid costly silicon respins.
This isn’t just about saving time and money, it’s also about enabling developers to move forward with confidence, knowing that the firmware foundation is solid even as the hardware continues to evolve.
Virtual platforms, especially fixed virtual platforms (FVPs), have become essential tools in firmware development for chiplet systems. These environments allow teams to emulate SoC behavior, validate firmware logic, and iterate on performance tuning without waiting for physical prototypes.
Integrated firmware stacks that support these platforms out of the box provide major advantages:
In most cases, this leads to first-pass success, where the initial silicon run boots and operates as expected.
As noted above, firmware isn’t just about booting the system anymore; it’s about managing it intelligently and proactively. In chiplet-based designs, where components may have independent power domains, trust zones, and interrupt schemes, firmware must provide a consistent abstraction layer that enables efficient communication and control. It must also incorporate predictive analytics to foresee potential system risks or failures and stop them before they become critical.
Optimized drivers and middleware ensure that surplus performance isn’t left on the table. Secure boot, telemetry, and power management are no longer optional, but now baseline requirements. Firmware must also enforce compliance with standards like ISO 26262 and IEC 61508, especially in safety-critical applications.
Remote management capabilities, such as OpenBMC integration and firmware management systems (FMS), extend this control across the lifecycle of the platform. These cannot be after-thoughts as they are of a holistic firmware strategy that supports both development and deployment, and they keep systems running for years in the field.
Chiplet silicon development is inherently collaborative; no single team owns the entire stack. Therefore, firmware must be designed to work across boundaries, with silicon providers, cloud service providers (CSPs), ODMs, and IP vendors all contributing to the final design.
This ecosystem-centric approach enables:
In this way, firmware becomes a shared language across the ecosystem, enabling faster innovation and closer alignment between hardware and software.
The technical benefits of modular firmware translate directly into measurable business outcomes:
In short, smartly designed firmware isn’t just a technical enabler. It’s a critical strategic asset.
Firmware development is entering a new phase, where automation and intelligence are starting to play a much larger role. AI-driven tools are being used to optimize performance, automate regression testing, and reduce cycle times.
These innovations are already delivering measurable savings, and they promise even greater agility for design teams working on the next generation of silicon processors. And as chiplet architecture continues to evolve, advanced firmware will remain central to their success.
In Conclusion, Chiplet-based silicon designs represent a major leap forward in custom silicon is designed, tested, built, and deployed at scale. But without a firmware foundation that’s equally modular, scalable, and intelligent, the benefits of chiplets can be lost in the integration complexity.
Integrating virtual designs with advanced firmware solutions provides the tools, validation capabilities, and ecosystem alignment needed to make chiplet systems possible. For developers working at the intersection of hardware and software, firmware is no longer a supporting element, it’s the system integrator, the performance enabler, and the security provider.
In simple terms, chiplets solve the silicon design scaling problem and firmware solves the integration problem. And with both working together, custom silicon just became a whole lot easier.
To learn more, please visit the Arm booth B11 at the 2025 OCP Global Summit , San Jose, CA, Oct 13-16 2025.
As a member of the Arm Total Design ecosystem and a leading Independent Firmware Vendor (IFV) with decades of experience, AMI is uniquely positioned to support Arm license partners through our AMI Developer Program for Arm Total Design. Learn more about the Program and get started on your chiplet design journey today by visiting https://www.ami.com/arm-solutions/.
All trademarks and registered trademarks in this publication are the property of their respective owners.