Scope

The International Conference on Parallel Architectures and Compilation Techniques (PACT) is a unique technical conference at the intersection of hardware and software, with a special emphasis on parallelism. PACT brings together researchers from computer architectures, compilers, execution environments, programming languages, and applications to present and discuss their latest research results, tools, and practical experiences. This year, PACT is specifically committed to pioneering AI-centric computing, seeking research that redefines the performance, scalability, and efficiency of large-scale AI workloads across diverse parallel and heterogeneous platforms.

PACT 2026 will be held as an in-person event in Chicago, IL. USA. We encourage all authors of accepted papers to participate, and at least one author must attend the conference.

PACT seeks submissions in two categories:

    Research Papers
    Tools and Practical Experience (TPE) Papers

Topics of Interest

PACT welcomes submissions on topics including, but not limited to:

    Parallel architectures, including accelerators for AI and other domains
        Conventional parallel architectures (e.g., multicore, multithreaded, superscalar, and VLIW architectures) and heterogeneous architectures
        AI accelerators: design of specialized hardware for LLM inference and training (e.g., TPUs, NPUs, and custom silicon)
        In-memory & near-data processing: architectures to mitigate the “memory wall” in massive AI model parameters
        Heterogeneous systems: integration of CPUs, GPUs, and FPGAs for distributed AI workloads
        Scalable AI Infrastructure: Architecture support for multi-node, multi-GPU clusters and high-speed interconnects for LLM scaling
    Compilers and tools for parallel architectures
        Conventional compilers and tools for parallel and heterogeneous architectures
        Dynamic translation and optimization
        ML compilers: automated optimization, kernel fusion, and code generation for ML frameworks
        LLMs for compilation: using AI to automate parallelization, loop transformations, and autotuning
        Dynamic optimization: runtime systems for adaptive AI model execution and sparse computation
        Quantization & compression: compiler-assisted techniques for model pruning and low-precision arithmetic
    Middleware and runtime system support for parallel computing
        Resource management & scheduling
        Communication & synchronization
        Energy-aware middleware
        Quantum-HPC interfacing
        Serverless parallel computing (e.g., AWS Lambda)
        AI & LLM-specific runtime support, including distributed inference & training, KV cache management, and computation-communication overlap
    I/O issues in parallel computing and their application impact
        Data loading & preprocessing pipelines
        Metadata scalability
        Memory-storage convergence
        Large-scale data processing for AI models and applications
    Hardware and software resilience & fault tolerance
        Checkpointing & restart
        Silent data corruption detection
        Self-healing runtimes
    Applications and experimental studies of parallel processing, especially using AI models
    Parallel programming languages, algorithms, and applications
    Computational models for concurrent execution
    Compiler and hardware support for parallel applications
    Support for correctness in hardware and software
    Reconfigurable parallel computing

Research Papers

Research papers will be evaluated by the PACT Program Committee based on:

    Relevance: The paper should align with PACT’s topics of interest.
    Novelty/Originality: The work should present new ideas or offer fresh perspectives.
    Significance: The research should address an important problem and have the potential to influence future work.
    Results: The claims should be well-supported by clear and validated results.
    Comparison to Prior Work: The paper should properly discuss existing literature, highlighting similarities, differences, and improvements.

Tools and Practical Experience (TPE) Papers

TPE papers focus on practical applications, industry challenges, and experience reports. A TPE paper must clearly explain its functionality, summarize practical experience with realistic case studies, and describe any supporting artifacts. The title of a TPE paper must include the prefix “TPE:”. TPE papers follow the same submission guidelines and are reviewed by the same Program Committee as research papers.

TPE papers will be evaluated based on:

    Originality: They should present PACT-related technologies applied to real-world problems.
    Usability: The tool or software should have broad applicability and aid PACT-related research.
    Documentation: The tool/software should be well-documented on a public website.
    Benchmark Repository: A benchmark suite should be provided for testing.
    Availability: Preference is given to tools/software that are freely available, though industry/commercial tools may be considered with justification.
    Foundations: The paper should relate to PACT’s principles, though extensive theoretical discussion is not required.