Automated Generation of Microfluidic Netlists using Large Language Models

Jasper Davidson; Skylar Stockham; Allen Boston; Ashton Snelgrove; Valerio Tenace; Pierre-Emmanuel Gaillardon

cs.AIFeb 22, 2026

Automated Generation of Microfluidic Netlists using Large Language Models

Authors: Jasper Davidson, Skylar Stockham, Allen Boston, Ashton Snelgrove, Valerio Tenace, Pierre-Emmanuel Gaillardon

Organizations: Department of Electrical and Computer Engineering, University of Utah · Primis AI, Inc.

Abstract

Microfluidic devices have emerged as powerful tools in various laboratory applications, but the complexity of their design limits accessibility for many practitioners. While progress has been made in microfluidic design automation (MFDA), a practical and intuitive solution is still needed to connect microfluidic practitioners with MFDA techniques. This work introduces the first practical application of large language models (LLMs) in this context, providing a preliminary demonstration. Building on prior research in hardware description language (HDL) code generation with LLMs, we propose an initial methodology to convert natural language microfluidic device specifications into system-level structural Verilog netlists. We demonstrate the feasibility of our approach by generating structural netlists for practical benchmarks representative of typical microfluidic designs with correct functional flow and an average syntactical accuracy of 88%.

Figures & tables

Fig. 1: Microfluidic netlist generation methodology using LLMs, spanning from laboratory applications to Verilog netlists that are compatible with existing MFDA techniques.

Model	Parameters	Layers	Context Length
Model	Parameters	Layers	(Number of Tokens)
CodeLlama	7 billion	32	16,384
Codestral	22 billion	56	32,768
Llama3	70 billion	32	8,192
Qwen2	72 billion	80	32,768
DeepSeek-R1	32 billion	60	131,072

TABLE I: Architectural Features of the Selected LLMs

	Benchmarks used for generating Prompting Methodology	Tests for:
1	Take 2 solutions as input. Mix them together.	Basic interpretation of prompts
2	Take 5 solutions as input to the experiment module. Mix the 5 solutions together parallel.	Reasoning in parallel
3	Take 5 solutions as input to the experiment module. Mix the 5 solutions together sequentially.	Reasoning in sequence
4	Take 3 solutions as input. Dilute the first solution, then mix with the other two solutions.	Reasoning with dilution
5	Put an algae solution into a heater. Mix the heated algae solution with itself to create a	Abstract reasoning
5	stirring effect. Pass through a membrane filter	Abstract reasoning

TABLE II: Microfluidic domain benchmark prompts utilized to develop and refine our LLM prompting methodology.

	Test Benchmarks	Tests for:
6	Take 2 solutions as input. Mix them together to create the output solution.	Basic interpretation of prompts
7	Take 4 solutions as input. Mix the 4 solutions together sequentially to create the output.	Reasoning in sequence
8	Take 6 solutions as input. Mix the 6 solutions together in parallel to create the output.	Reasoning in parallel
9	Take two solutions as input. Dilute the first solution, then mix with the other solution.	Reasoning with dilution
10	Heat up a solution of water. Filter the water to purify it, then mix with a diluted solution of oil.	Abstract Reasoning

TABLE III: Microfluidic domain benchmark prompts applied to assess and validate our LLM prompting methodology.

Fig. 2: A three-dimensional rendering of a microfluidic chip designed to mix six solutions in parallel, produced using OpenMFDA with a Qwen2-generated system of six mixers.

Specification	Verilog Syntax					Microfluidic Function
Pass@1[%]	CodeLlama	Codestral	Llama3	Qwen2	DeepDeek-R1	CodeLlama	Codestral	Llama3	Qwen2	DeepDeek-R1
Benchmark 6	80%	100%	100%	100%	100%	80%	100%	100%	100%	100%
Benchmark 7	80%	80%	100%	100%	100%	100%	100%	100%	100%	100%
Benchmark 8	0%	20%	60%	80%	100%	0%	40%	80%	100%	100%
Benchmark 9	80%	100%	60%	100%	80%	100%	100%	100%	100%	100%
Benchmark 10	0%	100%	60%	60%	60%	0%	100%	80%	100%	100%

TABLE IV: Pass@1 (single-short success rate) on MFD benchmarks testing for Verilog syntax and microfluidic functionality.

Program Synthesis

Program synthesis automatically constructs executable programs from examples, specifications, demonstrations, or natural-language descriptions.

See all 179 papers

Simulation Code Generation for Fluid Systems using Large Language Models: Benchmarking Models and Prompting Strategies
Jul 31, 2026Jan Marius Stürmer, Jascha Knack, Tobias Koch +1LLM Prompting Code Generation Evaluation
German Aerospace Centre (DLR), Institute for the Protection of Terrestrial Infrastructures, Rathausallee 12, 53757, Sankt Augustin, Germany · Algorithms for Computer Vision, Imaging and Data Analysis, Technische Hochschule Würzburg-Schweinfurt, Ignaz-Schön-Straße 11, Schweinfurt, 97421, Germany
Benchmarking LLMs for Verilog Design Flows
Jul 23, 2026Angshuman Chakravertty, Rahul Koshti, Buddhi Prakash Sharma +1Code Generation Code Generation Evaluation
School of Technology Management and Engineering, NMIMS, Hyderabad, India · Department of Electronics and Communication Engineering, Indian Institute of Technology, Roorkee, India · Department of Electrical and Electronics Engineering and the Anuradha and Prashanth Palakurthi Centre for Artificial Intelligence Research (APPCAIR), Birla Institute of Technology and Science, Pilani (BITS Pilani), Pilani 333031, India
Wrong Code, Right Structure: Learning Netlist Representations from Imperfect LLM-Generated RTL
Mar 10, 2026Siyang Cai, Cangyuan Li, Haoyu Gao +3Representation Learning Electronic Design Automation
School of Advanced Interdisciplinary Sciences, University of Chinese Academy of Sciences, Beijing, China · CICS, Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China · Hangzhou Institute for Advanced Study, University of Chinese Academy of Sciences, Hangzhou, China

Jul 31, 2026J/K move · Enter open · S save

Simulation Code Generation for Fluid Systems using Large Language Models: Benchmarking Models and Prompting Strategies

Jan Marius Stürmer, Jascha Knack, Tobias Koch, Andreas Weinmann

German Aerospace Centre (DLR), Institute for the Protection of Terrestrial Infrastructures, Rathausallee 12, 53757, Sankt Augustin, Germany · Algorithms for Computer Vision, Imaging and Data Analysis, Technische Hochschule Würzburg-Schweinfurt, Ignaz-Schön-Straße 11, Schweinfurt, 97421, Germany

LLM Prompting Code Generation Evaluation LLM Code Generation

Open paper AskPDF arXiv

Large language models (LLMs) have demonstrated a strong ability to generate syntactically correct code from natural-language specifications. In this study, we explore how LLMs can be harnessed to automatically translate a neutral graph representation of fluid system models into executable code for two widely adopted simulation environments: the Python library WNTR and the Modelica Standard Library. We conduct a systematic comparison of ten state-of-the-art LLMs and six prompting strategies that differ in the contextual information supplied (e.g., code or documentation). For each configuration we assess the generated code using a suite of software-quality metrics and we validate the functional fidelity of the resulting simulation models by reproducing benchmark fluid system scenarios. Our findings offer concrete guidance for researchers and engineers seeking to integrate LLM-driven code synthesis into model-based design pipelines. While the best-performing configurations achieve acceptable syntactic quality, we observe substantial gaps remain in simulation fidelity.

Abstract

Program Synthesis

Explore similar work

Simulation Code Generation for Fluid Systems using Large Language Models: Benchmarking Models and Prompting Strategies

Benchmarking LLMs for Verilog Design Flows

Wrong Code, Right Structure: Learning Netlist Representations from Imperfect LLM-Generated RTL