CircuitSmith

Analog circuit modeling library for MOSFETs and linear building blocks. Supports analytical equations, frequency response, parameter sweeps, and optimization.

Installation

uv venv
uv pip install -e .

For development:

uv pip install -e ".[dev]"
pre-commit install

Quick Start

from circuitsmith.blocks import CommonSourceAmplifier
from circuitsmith.plotting import plot_bode

cs = CommonSourceAmplifier(
    name="CS_Amp",
    gm=0.001,
    ro=100000,
    rd=10000,
    cl=1e-12,
    cgd=0.1e-12
)

metrics = cs.evaluate()
print(f"Gain: {metrics.dc_gain_db:.2f} dB")
print(f"Bandwidth: {metrics.bandwidth_hz:.2e} Hz")

plot_bode(cs, save_path="output/bode.png")

Run the demo:

circuitsmith demo

Available Blocks

• CommonSourceAmplifier, SourceFollower, CommonGate
• DifferentialPair, CurrentMirror
• SimpleTwoStageOpAmp
• RCLowPass, RLCLowPass, RLCBandPass

Examples

Ten runnable examples in examples/:

Frequency Analysis:

python examples/common_source_bode.py
python examples/cascaded_amplifier.py
python examples/two_stage_opamp_stability.py
python examples/rlc_filter_gallery.py

Parameter Studies:

python examples/source_follower_sweep.py
python examples/diffpair_cmrr.py
python examples/monte_carlo_gain.py

Transient Analysis:

python examples/transient_step_response.py
python examples/transient_amplifier_sine.py
python examples/transient_filter_noise.py

All plots save to output/.

CLI

circuitsmith demo
circuitsmith analyze common_source --gm 0.001 --ro 100000
circuitsmith sweep --param gm --values "0.001,0.002,0.004"

Analysis

Frequency Response:

from circuitsmith.analysis import FrequencyResponse
fr = FrequencyResponse.from_block(block, fmin=1e3, fmax=1e9)

Parameter Sweeps:

from circuitsmith.analysis import sweep_parameter
result = sweep_parameter(block, "gm", [0.001, 0.002, 0.004])

Monte Carlo:

from circuitsmith.analysis import monte_carlo
result = monte_carlo(block, {"gm": (0.001, 0.0001)}, n_samples=200)

Transient Analysis:

from circuitsmith.analysis import transient_step, transient_sine, transient_noisy_signal

step_resp = transient_step(block, step_amplitude=0.1, duration=1e-6)
sine_resp = transient_sine(block, frequency=1e6, amplitude=0.1)
noisy_resp = transient_noisy_signal(block, signal_frequency=1e6, noise_amplitude=0.2)

LLM Integration

CircuitSmith supports AI-powered circuit design using OpenAI GPT-4o-mini.

Install:

uv pip install -e ".[llm]"

Setup: Create a .env file in the project root:

OPENAI_API_KEY=your-api-key-here

Usage:

from circuitsmith.llm import OpenAICircuitDesignAgent

agent = OpenAICircuitDesignAgent()
spec = "Design a two-stage opamp with 60dB gain"
design = agent.design_from_spec(spec)

valid, errors = agent.validate_design(design)
if valid:
    print(f"Generated {len(design.blocks)} blocks")
    design.to_json()  # Export as JSON

Run the example:

python examples/llm_design_generation.py

Theory Model Disclaimer

CircuitSmith uses simplified first-order models suitable for:

• Design exploration
• Understanding tradeoffs
• Education

Not modeled:

• Short-channel effects
• Layout parasitics
• Nonlinear distortion
• Temperature and process corners beyond Monte Carlo

Always validate with SPICE using accurate models before tapeout.

Testing

pytest tests/ -v
ruff check .
mypy src/circuitsmith

License

MIT