kc-ml2 / meent

Electromagnetic simulation (RCWA) & optimization package in Python

Meent

Meent is an Electromagnetic(EM) simulation package with Python, composed of three main parts:

• Modeling
• EM simulation
• Optimization

Backends

Meent provides three libraries as a backend:

• NumPy
  • The fundamental package for scientific computing with Python
  • Easy and lean to use
• JAX
  • Autograd and XLA, brought together for high-performance machine learning research.
• PyTorch
  • A Python package that provides two high-level features: Tensor computation with strong GPU acceleration and Deep neural networks built on a tape-based autograd system

When to use

	Numpy	JAX	PyTorch	Description
64bit support	O	O	O	Default for scientific computing
32bit support	O	O	O	32bit (float32 and complex64) data type operation*
GPU support	X	O	O	except Eigendecomposition**
TPU support*	X	X	X	Currently there is no workaround to do 32 bit eigendecomposition on TPU
AD support	X	O	O	Automatic Differentiation (Back Propagation)
Parallelization	X	O	X	JAX pmap function

*In 32bit operation, operations on numbers of 8>= digit difference fail without warning or error. Use only when you do understand what you are doing.
**As of now(2023.03.19), GPU-native Eigendecomposition is not implemented in JAX and PyTorch. It's enforced to run on CPUs and send back to GPUs.

Numpy is simple and light to use. Suggested as a baseline with small ~ medium scale optics problem.
JAX and PyTorch is recommended for cases having large scale or optimization part.
If you want parallelized computing with multiple devices(e.g., GPUs), JAX is ready for that.
But since JAX does jit compilation, it takes much time at the first run.

How to install

pip install meent

JAX and PyTorch is needed for advanced utilization.

How to use

import meent

# backend 0 = Numpy
# backend 1 = JAX
# backend 2 = PyTorch

backend = 1
mee = meent.call_mee(backend=backend, ...)

Tutorials

Jupyter notebooks are prepared in tutorials to give a brief introduction.

Citation

To cite this repository:

@software{Kim_Meent_Electromagnetic_simulation,
  author = {Kim, Yongha and Kim, Sanmun and Lee, Jinmyoung and Jung, Anthony W. and Kim, Seolho},
  license = {MIT},
  title = {{Meent:Electromagnetic simulation & optimization package in Python}},
  url = {https://github.com/kc-ml2/meent}
}

Reference

Will be updated.

[1] https://opg.optica.org/josa/abstract.cfm?uri=josa-71-7-811, Rigorous coupled-wave analysis of planar-grating diffraction
[2] https://opg.optica.org/josaa/abstract.cfm?uri=josaa-12-5-1068
[3] https://opg.optica.org/josaa/abstract.cfm?uri=josaa-12-5-1077
[4] https://opg.optica.org/josaa/abstract.cfm?uri=josaa-13-5-1019
[5] https://opg.optica.org/josaa/abstract.cfm?uri=josaa-13-4-779
[6] https://opg.optica.org/josaa/abstract.cfm?uri=josaa-13-9-1870
[7] https://empossible.net/emp5337/
[8] https://github.com/zhaonat/Rigorous-Coupled-Wave-Analysis (see also our fork: https://github.com/yonghakim/zhaonat-rcwa)
[9] https://arxiv.org/abs/2101.00901

meent's Issues

modify 'setup.py' file

modify not to contain other folders than 'meent'

force wavelength to be float

optimizer structure

meent/meent/on_torch/optimizer.py

Lines 52 to 80 in 598d828

	class Optimizer:
	def __init__(self, optimizer, solver, target, lr = 0.001):
	self.solver = solver
	self.target = target
	if not isinstance(target, torch.Tensor):
	self.target = torch.Tensor(self.target)
	self.target.requires_grad = True
	if isinstance(optimizer,str):
	optimizer = getattr(torch.optim, optimizer)
	self.optimizer = optimizer([self.target], lr = lr)
	def optimize(self, iterations = 1000):
	for iteration in range(iterations):
	E_conv_all = to_conv_mat(self.solver.ucell, fourier_order)
	o_E_conv_all = to_conv_mat(1 / self.solver.ucell, fourier_order)
	de_ri, de_ti, _, _, _ = self.solver.solve(self.solver.wavelength, E_conv_all, o_E_conv_all)
	self.optimizer.zero_grad()
	loss = self.loss(de_ti)
	loss.backward()
	self.optimizer.step()
	print(loss)
	def loss(self, de_ti):
	return -de_ti[3, 2]

@yonghakim

Check and enforce Fourier_order type as integer

Even though Fourier_order was set to integer (e.g. 40) by user, it's changed to float (e.g. 40.0) and raises TypeError.