NL Net Foundation Funding Proposal PowerCommons A2O - OpenPower A2O Processor Revival Initiative

1. Project Overview and Expected Outcomes

Initially developed by IBM as part of the Blue Gene/Q supercomputing project, the OpenPower A2O processor represents a significant milestone in open-source computing architecture. This initiative aims to revive and modernize the A2O core, making it accessible for contemporary FPGA platforms and laying the foundation for future enterprise-grade computing, supercomputing solutions, and AI applications. Additionally, this is the fastest way to deliver a fully open-source and compliant CPU core built upon a proven architecture, decades of industry experience, and that can be easily modernized.

Technical Objectives: For this project, our primary goal is to restore full functionality to the A2O processor core, addressing build system incompatibilities, fixing broken build scripts with modern Vivado toolchains, addressing critical timing and synthesis issues, and eventually creating a robust software ecosystem around the processor. The project will deliver a working A2O implementation on modern Xilinx FPGA platforms, along with comprehensive documentation and a roadmap for ISA modernization and bringing the core in line with modern processor architectures (i.e., OpenPower ISA 3.1)

The A2O processor, in combination with a platform architecture that leverages the open-source and license-free MicroWatt OpenPOWER processor as the system boot processor and root of trust, can be utilized to create a fully open and verifiable SoC/system architecture. The system architecture aspect and MicroWatt-based platform architecture have been submitted in a companion proposal.

Current State and Challenges: The A2O processor code exists but suffers from bit rot - incompatible with current FPGA toolchains, undocumented build processes, and a lack of working software stacks. Our preliminary work has identified key technical barriers, including Vivado compatibility issues, missing peripheral implementations, and incomplete software and build toolchains. While some evidence exists that the core worked was barely functional on an FPGA board many years ago, it is unclear to what degree of functionality was tested and verified if at all. The comments and commit history in the OpenPower repository, paint a bleak picture.

Deliverables:

The working PowerCommons A2O core will be released in our public Git repository at:

2. Relevant Experience and Contributions

My involvement with OpenPower Foundation since May 2025 has provided hands-on expertise in OpenPower processor implementation, particularly through to the Microwatt project and broader ecosystem support.

Specific Technical Achievements

VCU118 Platform Enablement: I successfully added VCU118 FPGA board support to Microwatt SoC, enabling the processor to boot on this high-performance Xilinx platform. This required adapting the SoC infrastructure to the VCU118’s specific resources and constraints, including clock generation, I/O pin mappings, and resource utilization optimization.

DRAM Integration and Linux Boot: I further improved Microwatt Soc integration by adding support for VCU-118 DRAM controller transforming it from a demonstration core to a practical, Linux-capable system. This involved integrating DDR4 memory controllers, solving complex timing calibration issues. The successful Linux boot on VCU118 validated the complete hardware-software stack.

LiteX Framework Integration: While LiteX had limited Microwatt support, Linux boot was impossible as LiteX BIOS and bootstrapper only supported RISC-V. I enabled full Linux boot capability by::

• Adding PowerPC architecture support to the build system and BIOS
• Integrating new memory controller with proper timing calibration
• Resolving boot stack incompatibilities between LiteX and PowerPC
• Fixing interrupt controller memory region issues between Litex BIOS, Linux and Microwatt

These contributions established critical integration patterns that PowerCommons will extend for multi-core SoC development.

Community Infrastructure Support

I’ve assisted OpenPower community members in accessing multiple FPGA boards in remote environments, setting up:

• Board sharing protocols enabling 24/7 development across time zones
• Administering powercommons.org website and corresponding repositories
• Documentation and assisting new members

Open Source Contributions

My work has resulted in multiple upstream contributions:

• Microwatt repository: VCU118 platform support and DRAM integration
• LiteX repository: PowerPC architecture fixes and improvements
• Documentation: Setup guides and troubleshooting resources

You can refer to powercommons.org - I built and maintain the site along with build infrastructure and Git repositories.

This practical experience with real hardware, complex system integration, and community collaboration provides the foundation necessary to tackle PowerCommons’ ambitious goals of creating a fully open, verifiable SoC platform.

3. Budget Breakdown and Funding Utilization

We request a total funding of EUR 50,000. Note that the numbers below are indicative - given the highly complex nature of the project some funds may have to re-directed to secure equipment and/or licenses.

Budget Summary

Detailed Budget Breakdown

Personnel Costs (€40,000 - 80%)

The majority of the budget will cover daily expenses and compensation for project contributors. Given the specialized expertise required and budget constraints, the project will likely engage one primary contributor (myself), with the possibility of adding a second if a suitable candidate with deep technical knowledge can be identified at this rate. I might be the only contributor until additional funding can be secured.

Travel, Community and Dissemination (€7,000 - 14%)

Tools and Software (€2,000 - 4%)

Contingency and Administrative (€1,000 - 2%)

Task Breakdown by Phase

Additional Funding Sources

Currently, no other funding sources have been secured for this specific project. OpenPower Foundation provides in-kind support through expert consultation and community access. Future funding applications to the EU Horizon programs are planned for subsequent development phases.

The accelerated timeline requires higher daily rates to secure dedicated, experienced personnel capable of intensive development cycles within the compressed schedule.

4. Comparison with Historical Efforts

To the best of our knowledge, RISC-V and OpenPower are the only two widespread open source architectures. However, OpenPower and its parent Power Architectures have been around for much longer and are proven technologies. Additionally, OpenPower has a strict and well-established governance, compliance, and certification process, which makes it ideal for sensitive and highly secure environments.

In the long run, geopolitical considerations may influence the ecosystem, particularly given the geographic concentration of advanced RISC-V manufacturing and development capabilities. OpenPower Foundation, on the other hand, is a foundation with members from academia, industry, and the non-profit space from across the globe.

Greenfield vs Incremental

Previous OpenPower processor initiatives have predominantly been green field, individual-driven efforts, often constrained by single-person knowledge bottlenecks and limited institutional support. These projects attempted to design complete processors from scratch without fully leveraging existing, proven architectures.

Rather than starting from scratch, we build upon IBM’s proven A2O design - a member of the A2 family of processors that successfully powered Blue Gene/Q supercomputers. This foundation eliminates many fundamental design risks that plague ground-up processor projects. We aim to reuse and collaborate as much as possible, and deliver a working, open-core solution within a short timeframe. We then iterate and offer a modern version of the core in short sprints. We start with A2O, but we plan to integrate with and reuse the work that has already been delivered in projects like LiteX and LibreSoC, while complementing projects like PowerPC Notebook.

Our approach differs fundamentally through institutional backing from the OpenPower Foundation and direct guidance from Prof. Peter Hofstee, the original architect of IBM’s Cell Broadband Engine. This provides access to deep architectural knowledge that individual projects cannot replicate.

Key advantages include:

• Proven processor architecture with documented performance characteristics
• Direct access to original design expertise through Prof. Peter Hofstee
• OpenPower Foundation’s institutional support and community resources
• Focus on revival/modernization rather than complete redesign
• Intensive, professional development approach with compressed timeline
• Parallel effort on community building and involvement, working with the OpenPower Foundation

Formal verification opportunities

Unlike previous OpenPower processor initiatives that resulted in designs too complex for comprehensive formal verification, our approach creates unique opportunities for mathematical validation. MicroWatt, despite being a full-fledged 64-bit processor capable of booting Linux, maintains a remarkably small codebase (~20,000 lines of VHDL) that makes formal verification feasible. This compact yet powerful design enables us to pursue formal proofs of critical properties such as memory safety, instruction correctness, and absence of timing side-channels. By pairing the formally verifiable MicroWatt as a secure boot processor with the performance-oriented A2O core, PowerCommons can offer unprecedented security assurances—something impossible with proprietary processors or overly complex open designs like LibreSOC. This positions our project to deliver not just open hardware, but mathematically proven secure hardware, addressing critical infrastructure needs where verification is paramount.

5. Technical Challenges and Required Expertise

The A2O processor revival presents several significant technical challenges that require highly specialized expertise, combining multiple domains rarely found in a single individual.

FPGA Synthesis and Timing Closure: Modern FPGA toolchains have evolved significantly since A2O’s original implementation. Legacy Verilog code contains timing assumptions and synthesis directives incompatible with current Vivado versions. Resolving these requires a deep understanding of both processor microarchitecture and FPGA implementation strategies. Clock domain crossing, pipeline timing, and resource utilization optimization require expertise that bridges digital design and computer architecture.

Build System Modernization: The original A2O build environment uses obsolete scripts, dependency management, and compilation flows. Creating robust, reproducible build systems requires a combination of software engineering expertise and knowledge of hardware build tools. This includes makefile restructuring, dependency resolution, version control integration, and setting up continuous integration.

Software Toolchain Integration: Establishing working compiler toolchains, debuggers, and development tools requires expertise in compiler design, binary utilities, and processor instruction set architectures. The PowerPC instruction set implementation must be validated against processor behavior, requiring both software and hardware debugging capabilities.

System Integration and Peripheral Implementation: Integrating A2O with modern peripheral controllers, memory interfaces, and I/O systems requires expertise in systems engineering. This includes understanding bus protocols, interrupt handling, memory management units, and cache coherency implementations.

Linux Kernel Porting: Bringing up Linux on a revived A2O requires expertise in kernel internals, bootloader development, device tree configuration, and low-level system programming. Debugging kernel boot processes requires understanding both hardware behavior and operating system internals.

Accelerated Development Challenges: The compressed 6-month timeline intensifies these challenges, requiring rapid problem-solving capabilities and extensive parallel development workstreams. This requires exceptional project management skills, combined with technical expertise across all domains.

These challenges require a unique combination of computer architecture, FPGA design, system software, embedded programming, and intensive project management expertise—a skill set combination that is extremely rare and typically distributed across multiple specialists in industry settings.

6. Ecosystem Engagement and Deployment Strategy

Primary Ecosystem Actors

The OpenPower Foundation serves as our primary institutional partner, providing community access, technical guidance, and validation platforms. Their established relationships with hardware vendors, software developers, and research institutions create natural distribution channels for project outcomes.

Academic and research institutions are key targets for deployment, particularly those that require open-source, high-performance computing platforms for research and education. Universities with computer architecture programs benefit from accessible, well-documented processor implementations for teaching and research purposes.

Community Engagement Strategy

We will actively engage the broader OpenPower developer community through weekly technical updates, code contributions, and participation in foundation meetings. Monthly progress reports and quarterly technical demonstrations will maintain community visibility and gather feedback throughout the accelerated development cycle.

The FPGA development community, particularly those focused on open-source designs, represents another critical engagement target.

Industry Partnerships

Collaboration with FPGA board manufacturers ensures the availability of reference implementations on accessible development platforms. This includes partnerships with companies such as Digilent, Terasic, and others, which provide educational and development boards.

The retro computing community offers immediate deployment opportunities through specialized gaming and hobby computing applications. These users provide early adoption feedback and potential revenue streams, which are crucial for project sustainability.