AnalogLib — Analog In-Memory Computing Library

AnalogLib is an open-source Python library for simulating analog in-memory computing (IMC) and neural network inference on resistive crossbar architectures (ReRAM, PCM, and memristive arrays).

It models Ohm's Law ($I = V \cdot G$) and Kirchhoff's Current Law array execution, mapping high-level ML models (NumPy, PyTorch) onto physical crossbar arrays while accounting for device state quantization, read noise, wire IR drop, thermal scaling, retention drift, and peripheral ADC/DAC converters.

---

⚡ 5-Minute Quickstart

import analoglib as al
import numpy as np

# 1. Define model weights
W1 = np.random.uniform(-0.5, 0.5, (128, 64))
W2 = np.random.uniform(-0.5, 0.5, (64, 10))

# 2. Build AnalogModel via AIR (Analog Intermediate Representation)
model = al.AnalogModel.from_numpy([W1, W2], activations=["relu", "softmax"])

# 3. Target physical ReRAM crossbars + 8-bit peripherals
model.compile(
    device=al.ReRAM(g_min=1e-6, g_max=100e-6, num_states=256, read_noise_sigma=0.01),
    adc_bits=8,
    dac_bits=8,
)

# 4. Simulate hardware inference
x_input = np.random.uniform(0.0, 1.0, 128)
result = model.simulate(x_input, mode="hardware")

# 5. Print hardware metrics report
result.report()

---

📊 v0.1.0 Capability Matrix

Component	Supported in v0.1.0	Planned / Roadmap
Frameworks	✅ NumPy, ✅ PyTorch (nn.Module)	🚧 TensorFlow / Keras (v0.2.0), 🚧 ONNX (v0.3.0)
Device Models	✅ IdealDevice, ✅ ReRAM	🚧 Non-linear J4 memristor model
Mapping	✅ Differential ($G^+, G^-$), ✅ Offset ($G_{\text{mid}}$)	🚧 Multi-bit cell slicing
Array Architecture	✅ Crossbar, ✅ TiledCrossbar	🚧 3D Crossbar stack
Peripherals	✅ DAC, ✅ ADC	🚧 Non-uniform logarithmic ADC
Simulation Modes	✅ "ideal", ✅ "device", ✅ "hardware"	🚧 Transient pulse-based simulation
Physical Effects	✅ IRDrop, ✅ Thermal, ✅ Drift	🚧 MNA full nodal solver
Exporters & Format	✅ Encrypted .analog, ✅ SPICE (ngspice/LTspice)	🚧 Verilog-A behavioral model export
Analytics & CLI	✅ AnalogProfiler, ✅ analog CLI	🚧 Graphical dashboard UI

---

📦 Installation

# Base package
pip install analoglib

# With PyTorch support
pip install "analoglib[torch]"

# With Visualization support
pip install "analoglib[viz]"

# Full developer installation
pip install "analoglib[all]"

---

📖 Complete Documentation System

Explore the full documentation system:

• 🚀 Getting Started — Installation, Quickstart, First VMM, Core Concepts
• 🧠 Physics & Equations — Conductance mapping, state quantization, device noise, hardware modes
• 🔬 Capabilities Matrix — Full v0.1.0 compatibility breakdown
• ❓ FAQ & AI Discoverability — Answers to PyTorch, Keras, VMM, SPICE, and ReRAM questions
• ⚡ Framework Converters — PyTorch (from_torch) & NumPy (from_numpy) guides
• 🔒 File Format Spec — Encrypted .analog binary format layout
• 🔌 SPICE Exporter — ngspice & LTspice netlist generation
• 📊 Analytics & Profiler — TOPS/W, power, energy, area estimation
• 🛠️ Troubleshooting Reference — Error codes and fix guide
• 📚 API Reference Index — Complete module, class, and method index
• 🗺️ Release Roadmap — Feature roadmap for v0.2.0 and beyond

---

📜 License

AnalogLib is released under the MIT License.