EcoLogic 🌍

Real-Time, Energy-Efficient AI with DiffLogic on FPGA

EcoLogic is a neurosymbolic AI solution that leverages the DiffLogic architecture on an FPGA to achieve high-speed, low-power inference, ideal for real-time applications like disaster response. EcoLogic combines the efficiency of FPGA technology with IBM’s watsonx.ai platform to provide a scalable, sustainable AI model that addresses the energy demands of data centers and supports climate goals.

---

Features ⚡️

• DiffLogic Model Training: Optimized model training with differentiable logic gates, enabling faster inference while maintaining a sparse, energy-efficient structure.
• FPGA Implementation: First-of-its-kind hardware adaptation of DiffLogic on the DE10-Lite FPGA for even faster inference.
• IBM watsonx.ai Integration: Utilizes IBM’s generative AI platform for real-time analysis in AI-driven disaster response.

---

Repository Structure 📁

• ModelSim/: Contains the latest testbench HDL files for simulation.
• Quartus/: Houses the most recent Quartus project files for FPGA synthesis and implementation.
• config/: Configuration files, including support for 16x16 image inputs, allowing flexible input sizes.
• trained_models/: Pre-trained DiffLogic models ready for FPGA deployment.
• verilog/: Verilog files for model deployment on the FPGA.
• vhdl/: VHDL files corresponding to the Verilog implementations, providing hardware flexibility.
• watson ai/: Scripts for IBM watsonx.ai integration, including data generation and DAD (Data Analyzer and Debugger) logic analyzer.

---

Setup and Installation 🚀

1. Clone the Repository

   git clone https://github.com/matheusmaldaner/EcoLogic.git
   cd EcoLogic

2. Install Dependencies
   Install the required Python packages:

   pip install -r requirements.txt

2.1 Install DiffLogic Refer to DiffLogic's official implementation for directions on how to install it.

3. FPGA Synthesis
   Open the Quartus project file (.qpf) in the Quartus/ directory and follow the instructions in the README.md file there to compile and deploy to the DE10-Lite.

---

Usage 💻

1. Model Training
   Run Main.ipynb to train DiffLogic models on your dataset. The notebook allows customization of hyperparameters and visualizes training metrics.

2. Hardware Implementation
   The verilog/ and vhdl/ directories contain files for deploying trained models onto the FPGA. Use ModelSim and Quartus for simulation and synthesis.

3. IBM watsonx.ai Integration
   The watson ai/ directory includes a script for watsonx.ai data analysis and inference testing. Run the integration script to connect EcoLogic with IBM’s platform.

---

Example Input and Datasets 📊

• mnist_dataset.py: Script to prepare and preprocess the MNIST dataset for DiffLogic training.
• mnist_input.mif: Memory initialization file for FPGA ROM setup.

---

Results and Benchmarks 📈

Nodes	DiffLogic (µs)	FPGA (µs)	% Decrease
512000	365.7	139.46	62%
3584000	537.89	177.9	67%
6144000	724.20	186.76	74%
8704000	907.78	202.54	78%
11264000	1086.52	239.18	78%

---

Contributing 🤝

Contributions are welcome! Please fork this repository and submit a pull request. Ensure that your code passes all tests and adheres to our style guide.