def runSim(params, eps_bg, eps_wg, x_pos, y_pos, size_x, filter_R, beta_start=1): ######## DEFINE A 2D TOPOLOGY OPTIMIZATION REGION ######## geometry = TopologyOptimization2D(params=params, eps_min=eps_bg, eps_max=eps_wg, x=x_pos, y=y_pos, z=0, filter_R=filter_R, beta=beta_start) ######## DEFINE FIGURE OF MERIT ######## # The base simulation script defines a field monitor named 'fom' at the point where we want to modematch to the fundamental TE mode fom = ModeMatch(monitor_name='fom', mode_number='Fundamental TE mode', direction='Forward', norm_p=2) ######## DEFINE OPTIMIZATION ALGORITHM ######## optimizer = ScipyOptimizers(max_iter=50, method='L-BFGS-B', scaling_factor=1, pgtol=1e-6, ftol=1e-4, target_fom=0.5, scale_initial_gradient_to=0.25) ######## LOAD TEMPLATE SCRIPT AND SUBSTITUTE PARAMETERS ######## script = load_from_lsf( os.path.join(CONFIG['root'], 'examples/Ysplitter/splitter_base_2D_TE_topology.lsf')) script = script.replace('opt_size_x=3.5e-6', 'opt_size_x={:1.6g}'.format(size_x)) wavelengths = Wavelengths(start=1450e-9, stop=1650e-9, points=11) opt = Optimization(base_script=script, wavelengths=wavelengths, fom=fom, geometry=geometry, optimizer=optimizer, use_deps=False, hide_fdtd_cad=True, plot_history=False, store_all_simulations=False) ######## RUN THE OPTIMIZER ######## opt.run()
def test_shape_boundary_approximation_in_3D(self): print( "3D optimization with shape boundary approximation (use_deps = False): " ) self.geometry.bounds = [(0.0, self.wg_width / 2.0 - self.mesh_del) ] * len(self.geometry.bounds) # Note: bounds are tweaked since the shape boundary approximation method does not work # when the shape under optimization touches the boundary of the FDTD region. opt = Optimization(base_script=self.base_script + "setnamed('FDTD','dimension','3D');", wavelengths=self.wavelengths, fom=self.fom, geometry=self.geometry, optimizer=self.optimizer, hide_fdtd_cad=True, use_deps=False, plot_history=False, store_all_simulations=False) fom, params = opt.run() self.assertGreaterEqual(fom, 0.972) self.assertAlmostEqual(params[0], (self.wg_width / 2.0 - self.mesh_del) * self.optimizer.scaling_factor) self.assertAlmostEqual(params[1], (self.wg_width / 2.0 - self.mesh_del) * self.optimizer.scaling_factor)
def test_shape_boundary_approximation(self): print( "varFDTD optimization with shape boundary approximation (use_deps = False): " ) self.geometry.bounds = [ (self.mesh_del, self.wg_width / 2.0 - self.mesh_del) ] * len(self.geometry.bounds) opt = Optimization(base_script=self.base_script, wavelengths=self.wavelengths, fom=self.fom, geometry=self.geometry, optimizer=self.optimizer, use_var_fdtd=True, hide_fdtd_cad=True, use_deps=False, plot_history=False, store_all_simulations=False) fom, params = opt.run() self.assertGreaterEqual(fom, 0.972) self.assertAlmostEqual(params[0], (self.wg_width / 2.0 - self.mesh_del) * self.optimizer.scaling_factor) self.assertAlmostEqual(params[1], (self.wg_width / 2.0 - self.mesh_del) * self.optimizer.scaling_factor)
def test_broadband(self): print("Broadband optimization results:") opt = Optimization(base_script = self.base_script, wavelengths = self.wavelengths, fom = self.fom, geometry = self.geometry, optimizer = self.optimizer, hide_fdtd_cad = True, use_deps = True) fom, params = opt.run() self.assertAlmostEqual(params[0], 2.050400e-7 * self.optimizer.scaling_factor, 5) self.assertGreaterEqual(fom, 0.4618)
def test_broadband_optimization(self): print("Broadband optimization results (use_deps = True):") opt = Optimization(base_script=self.base_script, wavelengths=self.wavelengths, fom=self.fom, geometry=self.geometry, optimizer=self.optimizer, use_var_fdtd=False, hide_fdtd_cad=True, use_deps=True, plot_history=False, store_all_simulations=False) fom, params = opt.run() self.assertAlmostEqual(params[0], 2.050375e-7 * self.optimizer.scaling_factor[0], 4) self.assertGreaterEqual(fom, 0.461815)
def test_single_frequency(self): print("Single frequency optimization results:") self.fom.target_T_fwd = lambda wl: np.ones(wl.size) self.fom.multi_freq_src = False self.wavelengths = 1550.0e-9 self.optimizer.scaling_factor = 2.0e7 self.optimizer.pgtol = 3.1e-2 opt = Optimization(base_script = self.base_script, wavelengths = self.wavelengths, fom = self.fom, geometry = self.geometry, optimizer = self.optimizer, hide_fdtd_cad = True, use_deps = True) fom, params = opt.run() self.assertAlmostEqual(params[0], 2.058006e-7 * self.optimizer.scaling_factor, 5) self.assertGreaterEqual(fom, 0.9192)
def test_permittivity_derivatives_in_2D(self): print( "2D optimization with permittivity derivatives (use_deps = True): " ) opt = Optimization(base_script=self.base_script + "setnamed('FDTD','dimension','2D');", wavelengths=self.wavelengths, fom=self.fom, geometry=self.geometry, optimizer=self.optimizer, hide_fdtd_cad=True, use_deps=True) fom, params = opt.run() self.assertGreaterEqual(fom, 0.99991) self.assertAlmostEqual( params[0], self.wg_width / 2.0 * self.optimizer.scaling_factor) self.assertAlmostEqual( params[1], self.wg_width / 2.0 * self.optimizer.scaling_factor)
def test_permittivity_derivatives(self): print( "varFDTD optimization with permittivity derivatives (use_deps = True): " ) opt = Optimization(base_script=self.base_script, wavelengths=self.wavelengths, fom=self.fom, geometry=self.geometry, optimizer=self.optimizer, use_var_fdtd=True, hide_fdtd_cad=True, use_deps=True, plot_history=False, store_all_simulations=False) fom, params = opt.run() self.assertGreaterEqual(fom, 0.99991) self.assertAlmostEqual( params[0], self.wg_width / 2.0 * self.optimizer.scaling_factor) self.assertAlmostEqual( params[1], self.wg_width / 2.0 * self.optimizer.scaling_factor)
def test_single_wavelength_legacy_optimization(self): print("Single wavelength optimization results (use_deps = False):") self.fom.target_T_fwd = lambda wl: np.ones(wl.size) self.fom.multi_freq_src = False self.wavelengths = 1550.0e-9 self.optimizer.scaling_factor = np.array(2.0e7) self.optimizer.pgtol = 3.1e-2 opt = Optimization(base_script=self.base_script, wavelengths=self.wavelengths, fom=self.fom, geometry=self.geometry, optimizer=self.optimizer, use_var_fdtd=False, hide_fdtd_cad=True, use_deps=False, plot_history=False, store_all_simulations=False) fom, params = opt.run() self.assertAlmostEqual(params[0], 2.05609116e-7 * self.optimizer.scaling_factor, 4) self.assertGreaterEqual(fom, 0.91905)
depth=wg_height, eps_out=1.0**2, eps_in=3.47668**2, edge_precision=5, dx=1.0e-5) ######## DEFINE FIGURE OF MERIT ######## fom = ModeMatch(monitor_name='fom', mode_number=3, direction='Backward', target_T_fwd=lambda wl: 0.5 * np.ones(wl.size), norm_p=1) ######## DEFINE OPTIMIZATION ALGORITHM ######## optimizer = ScipyOptimizers(max_iter=200, method='L-BFGS-B', scaling_factor=1.0, pgtol=1.0e-4) ######## PUT EVERYTHING TOGETHER ######## opt = Optimization(base_script=base_sim, wavelengths=wavelengths, fom=fom, geometry=geometry, optimizer=optimizer, hide_fdtd_cad=False, use_deps=True) ######## RUN THE OPTIMIZER ######## opt.run()
def runGratingOptimization(bandwidth_in_nm, etch_depth_shallow, etch_depth_deep, n_grates, initial_params = None): ### Yet another parametrization which allows to enforce minimum feature size when the optimizer only supports box constraints ### params = [x0, a1, b1, ..., aN] if initial_params is None: params = np.zeros(4*n_grates) for i in range(n_grates): params[i*4] = 0.2 #< Width up params[i*4+1] = 0.4*(i/n_grates) #< Width of the shallow etch params[i*4+2] = 0.1 #< Width up params[i*4+3] = 0.4*(i/n_grates) #< Width of the deep etch params[0] = 0 #< Overwrite the first since it has a special meaning: Start of the grating at 0um else: params = initial_params bounds = [(0, 1)]*(4*n_grates) bounds[0] = (-3,3) def grating_params_pos(params, output_waveguide_length = 0.5e-6, height = 220e-9, y0 = 0): x_begin = -3e-6 y3 = y0+height y2 = y3-etch_depth_deep y1 = y3-etch_depth_shallow x0 = params[0]*1e-6 #< First parameter is the starting position verts = np.array( [ [x_begin,y0],[x_begin,y3],[x0,y3],[x0,y1] ] ) ## Iterate over all but the last for i in range(n_grates-1): x1 = x0 + params[i*4+1]*1e-6 #< Width of the deep etch x2 = x1 + params[i*4+2]*1e-6 #< Width up x3 = x2 + params[i*4+3]*1e-6 #< Width of the shallow etch x4 = x3 + params[i*4+4]*1e-6 #< Width up verts = np.concatenate((verts,[[x1,y1],[x1,y3],[x2,y3],[x2,y2],[x3,y2],[x3,y3],[x4,y3],[x4,y1]]),axis=0) x0 = x4 x1 = x0 + params[(n_grates-1)*4+1]*1e-6 #< Width of the deep etch x2 = x1 + params[(n_grates-1)*4+2]*1e-6 #< Width up x3 = x2 + params[(n_grates-1)*4+3]*1e-6 #< Width of the shallow etch x_end = x3+output_waveguide_length verts = np.concatenate((verts,[[x1,y1],[x1,y3],[x2,y3],[x2,y2],[x3,y2],[x3,y3],[x_end,y3],[x_end,y0]]),axis=0) return verts geometry = FunctionDefinedPolygon(func = grating_params_pos, initial_params = params, bounds = bounds, z = 0.0, depth = 220e-9, eps_out = 1.44 ** 2, eps_in = 3.47668 ** 2, edge_precision = 5, dx = 1e-3) ######## DEFINE FIGURE OF MERIT ######## fom = ModeMatch(monitor_name = 'fom', mode_number = 1, direction = 'Backward', target_T_fwd = lambda wl: np.ones(wl.size), norm_p = 1) ######## DEFINE OPTIMIZATION ALGORITHM ######## optimizer = ScipyOptimizers(max_iter = 250, method = 'L-BFGS-B', scaling_factor = 1, pgtol = 1e-6) #SLSQP ######## DEFINE BASE SIMULATION ######## base_script = load_from_lsf(os.path.join(os.path.dirname(__file__), 'grating_coupler_2D_2etch.lsf')) ######## PUT EVERYTHING TOGETHER ######## lambda_start = 1550 - bandwidth_in_nm/2 lambda_end = 1550 + bandwidth_in_nm/2 lambda_pts = int(bandwidth_in_nm/10)+1 wavelengths = Wavelengths(start = lambda_start*1e-9, stop = lambda_end*1e-9, points = lambda_pts) opt = Optimization(base_script = base_script, wavelengths = wavelengths, fom = fom, geometry = geometry, optimizer = optimizer, hide_fdtd_cad = True, use_deps = True) ######## RUN THE OPTIMIZER ######## opt.run()
def runGratingOptimization(bandwidth_in_nm, etch_depth, n_grates, params): bounds = [(0.1, 1)] * 4 bounds[0] = (-3, 3) #< Starting position bounds[1] = (0, 0.1) #< Scaling parameter R bounds[2] = (1.5, 3) #< Parameter a bounds[3] = (0, 2) #< Parameter b def grating_params_pos(params, output_waveguide_length=0.5e-6, height=220e-9, y0=0): x_begin = -3e-6 y3 = y0 + height y1 = y3 - etch_depth x_start = params[0] * 1e-6 #< First parameter is the starting position x0 = x_start R = params[1] * 1e6 #< second parameter (unit is 1/um) a = params[2] #< Third parameter (dim-less) b = params[3] #< Fourth parameter (dim-less) verts = np.array([[x_begin, y0], [x_begin, y3], [x0, y3], [x0, y1]]) lambda_c = 1.55e-6 F0 = 0.95 ## Iterate over all but the last for i in range(n_grates - 1): F = F0 - R * (x0 - x_start) Lambda = lambda_c / (a + F * b) x1 = x0 + (1 - F) * Lambda #< Width of the etched region x2 = x0 + Lambda #< Rest of cell verts = np.concatenate( (verts, [[x1, y1], [x1, y3], [x2, y3], [x2, y1]]), axis=0) x0 = x2 F = F0 - R * (x0 - x_start) Lambda = lambda_c / (a + F * b) x1 = x0 + (1 - F) * Lambda #< Width of the etched region x_end = x1 + output_waveguide_length verts = np.concatenate( (verts, [[x1, y1], [x1, y3], [x_end, y3], [x_end, y0]]), axis=0) return verts geometry = FunctionDefinedPolygon(func=grating_params_pos, initial_params=params, bounds=bounds, z=0.0, depth=110e-9, eps_out=1.44**2, eps_in=3.47668**2, edge_precision=5, dx=1e-3) ######## DEFINE FIGURE OF MERIT ######## fom = ModeMatch(monitor_name='fom', mode_number=1, direction='Backward', target_T_fwd=lambda wl: np.ones(wl.size), norm_p=1) ######## DEFINE OPTIMIZATION ALGORITHM ######## optimizer = ScipyOptimizers(max_iter=25, method='L-BFGS-B', scaling_factor=1, pgtol=1e-6) ######## DEFINE BASE SIMULATION ######## base_script = load_from_lsf( os.path.join(os.path.dirname(__file__), 'grating_coupler_2D_2etch.lsf')) ######## PUT EVERYTHING TOGETHER ######## lambda_start = 1550 - bandwidth_in_nm / 2 lambda_end = 1550 + bandwidth_in_nm / 2 lambda_pts = int(bandwidth_in_nm / 10) + 1 wavelengths = Wavelengths(start=lambda_start * 1e-9, stop=lambda_end * 1e-9, points=lambda_pts) opt = Optimization(base_script=base_script, wavelengths=wavelengths, fom=fom, geometry=geometry, optimizer=optimizer, hide_fdtd_cad=True, use_deps=True) ######## RUN THE OPTIMIZER ######## opt.run()
class LumericalInverseDesign: """ Wrapper for Lumerical lumopt (part of it) Parameters ---------- max_iter: int method: string scaling_factor: float pgtol: float ftol: float wavelength_start: float wavelength_stop: float wavelength_points: float build_simulation: string fom: obj geometry: obj hide_fdtd_cad: bool """ def __init__(self, max_iter, method, scaling_factor, pgtol, ftol, wavelength_start, wavelength_stop, wavelength_points, build_simulation, fom, geometry, hide_fdtd_cad): # The optimizer must be generated anew at each iteration self._new_local_optimizer = ScipyOptimizers( max_iter=max_iter, method=method, scaling_factor=scaling_factor, ftol=ftol, pgtol=pgtol) self._wl = Wavelengths(start=wavelength_start, stop=wavelength_stop, points=wavelength_points) self._optimization = Optimization(base_script=build_simulation, wavelengths=self._wl, fom=fom, geometry=geometry, optimizer=self._new_local_optimizer, hide_fdtd_cad=hide_fdtd_cad, use_deps=True) def run(self): """ Run the lumopt optimization Returns ------- res: array The figure of merit of the optimized device param: numpy array The optimized parameters """ results = self._optimization.run() self._optimization.sim.fdtd.close() # plot optimization recap figure plt.show() return [results[0], np.array(results[1])] def _cleanup(self): ''' Remove all the folders generated by lumopt ''' # folder for the file local_folder = os.path.realpath( os.path.join(os.getcwd(), os.path.dirname(__file__))) subdir_list = os.listdir(local_folder) for folder in subdir_list: if folder.startswith('opts_') or (folder.startswith('optimization') and folder.endswith('.png')): shutil.rmtree(local_folder + '\\' + folder, ignore_errors=True) def __del__(self): # Remove objects to delete pointers or pickle could have problems del self._optimization del self._wl del self._new_local_optimizer self._cleanup()