Abstract:High-Level Synthesis (HLS) tools offer rapid hardware design from C code, but their compatibility is limited by code constructs. This paper investigates Large Language Models (LLMs) for automatically refactoring C code into HLS-compatible formats. We present a case study using an LLM to rewrite C code for NIST 800-22 randomness tests, a QuickSort algorithm, and AES-128 into HLS-synthesizable C. The LLM iteratively transforms the C code guided by the, implementing functions like streaming data and hardware-specific signals. With the hindsight obtained from the case study, we implement a fully automated framework to refactor C code into HLS-compatible formats using LLMs. To tackle complex designs, we implement a preprocessing step that breaks down the hierarchy in order to approach the problem in a divide-and-conquer bottom-up way. We validated our framework on three ciphers, one hash function, five NIST 800-22 randomness tests, and a QuickSort algorithm. Our results show a high success rate on benchmarks that are orders of magnitude more complex than what has been achieved generating Verilog with LLMs.

What problem does this paper attempt to address?

### Problems the paper attempts to solve This paper aims to solve the compatibility problems in the conversion process from C code to Hardware Description Language (HDL). Specifically, the paper explores how to use large - language models (LLMs) to automatically refactor C code so that it can be compatible with High - Level Synthesis (HLS) tools. HLS tools can quickly generate hardware designs from C code, but their compatibility with C code is limited, especially for certain specific code structures. This results in the need for manual adjustment and refactoring of C code to meet the requirements of HLS tools in practical applications, which is both time - consuming and error - prone. ### Main contributions 1. **Automation process**: - **Support hierarchical design**: Automatically build unit tests from top - level tests. - **Function rewriting**: Support the rewriting of streaming functions. - **Generate optimization instructions**: Support the generation of optimization instructions (such as pragmas) for area or throughput targets. - **Cost control**: Use in - context learning and intelligent model selection to reduce costs. 2. **Experimental evaluation**: - Use real - world application benchmarks for evaluation, including the implementation of encryption algorithms and NIST randomness tests. ### Method overview 1. **Background introduction**: - **High - Level Synthesis (HLS)**: Introduces the capabilities and limitations of HLS tools, including unsupported features such as dynamic memory allocation, recursion, and pointers. - **Large - language models (LLMs)**: Discusses the advantages of LLMs in code generation and translation, especially in widely used programming languages such as C, C++, and Python. 2. **Related work**: - Outlines previous work on using LLMs to design hardware, including Verigen, ChipChat, AutoChip, etc. - Emphasizes the uniqueness of the method in this paper, that is, using HLS tools by generating synthesis - compatible C code instead of directly generating HDL code. 3. **Case study**: - **Interactive method**: Use the Gemini LLM to refactor the C code of NIST 800 - 22 randomness tests, QuickSort algorithm, and AES - 128 encryption to make it suitable for HLS tools. - **Methodology**: Details each step, including removing print statements, rewriting streaming interfaces, optimizing data types, etc. - **Results**: Shows the performance of LLM in different tasks, including successful and failed cases. 4. **Fully automated C2HLSC framework**: - **Pre - processing step**: Handle hierarchical design and decompose multi - function tasks into single - function tasks. - **Dual - feedback loop**: Ensure that the generated code can be both compiled successfully and synthesized successfully by HLS tools. - **Optimization instruction generation**: After the code is synthesizable, apply optimization instructions (such as pragmas). ### Conclusion This paper significantly reduces the time and effort required for manual code adjustment to adapt to HLS tools by using LLMs to automatically refactor C code. The experimental results show that this method has a high success rate in handling complex designs, demonstrating the great potential of LLMs in hardware design.

C2HLSC: Leveraging Large Language Models to Bridge the Software-to-Hardware Design Gap

C2HLSC: Can LLMs Bridge the Software-to-Hardware Design Gap?

Are LLMs Any Good for High-Level Synthesis?

Evaluating Large Language Models for Automatic Register Transfer Logic Generation via High-Level Synthesis

Optimizing High-Level Synthesis Designs with Retrieval-Augmented Large Language Models

HLSPilot: LLM-based High-Level Synthesis

Automated C/C++ Program Repair for High-Level Synthesis via Large Language Models

Evaluating LLMs for Hardware Design and Test

Advanced Large Language Model (LLM)-Driven Verilog Development: Enhancing Power, Performance, and Area Optimization in Code Synthesis

LLM-Aided Efficient Hardware Design Automation

Digital ASIC Design with Ongoing LLMs: Strategies and Prospects

From English to ASIC: Hardware Implementation with Large Language Model

Leveraging High-Level Synthesis and Large Language Models to Generate, Simulate, and Deploy a Uniform Random Number Generator Hardware Design

ChatChisel: Enabling Agile Hardware Design with Large Language Models

Using LLM such as ChatGPT for Designing and Implementing a RISC Processor: Execution,Challenges and Limitations

Benchmarking Large Language Models for Automated Verilog RTL Code Generation

Large Language Models to Generate System-Level Test Programs Targeting Non-functional Properties

Make Every Move Count: LLM-based High-Quality RTL Code Generation Using MCTS

LLM4SecHW: Leveraging Domain Specific Large Language Model for Hardware Debugging

HPC-Coder-V2: Studying Code LLMs Across Low-Resource Parallel Languages

Fixing Hardware Security Bugs with Large Language Models