Merge branch 'main' of https://github.com/ChrisZonghaoLi/sky130_ldo_rl …

…into main

README.md

@@ -4,18 +4,18 @@
 
 ---
 
-This is the source code for the paper submitted to IEEE/ACM ICCAAD 2023.
+This is the source code for the paper accepted to IEEE/ACM ICCAAD 2023. The final accepted manuscript can be found here: https://www.zonghaoli.com/ml_analog_ic.html.
 
 ---
 
 ## Repo Structure
 
-All files you are lookring for are placed inside the folder `/python`. 
+All files you are looking for are placed inside the folder `/python`. 
 - `ldo.py`: define the LDO1 environment (Gymnasium compatible).
 - `ldo_folded_cascode.py`: define the LDO2 environment (Gymnasium compatible).
 - `ckt_graphs.py`: define the graph info as well as specifications for LDO1 and LDO2.
-- `dev_params.py`: used to extrac device parameters such as threshold voltage and transconductance of transistors, providing the observations for RL.
-- `ddpg.py`: where DDPG algorithm is stored.
+- `dev_params.py`: used to extract device parameters such as threshold voltage and transconductance of transistors, providing the observations for RL.
+- `ddpg.py`: where the DDPG algorithm is stored.
 - `models.py`: where various GNN models are stored.
 - `main.py`: run the optimization for LDO1.
 - `main2.py`: run the optimization for LDO2.
@@ -25,4 +25,7 @@ All files you are lookring for are placed inside the folder `/python`.
 - `/simulations`: store the SPICE files and it is where Ngspice is running.
 - `ldo_rgcn_rl.ipynb`: a notebook summaries the work, it is outdated.
 
-
+---
+## Getting Started
+You may just follow the tutorial in `ldo_rgcn_rl.ipynb`, which is outdated but the flow is the same.
+If you want to have the update-to-date version running, simply execute `main.py` for LDO1 and `main2.py`for LDO2. 

python/main.py

@@ -105,7 +105,7 @@
     save_name_actor = f"Actor_{CktGraph().__class__.__name__}_{date}_noise={noise_type}_reward={best_reward:.2f}_{GNN().__class__.__name__}_rew_eng={rew_eng}.pth"
 
     model_weight_critic = agent.critic.state_dict()
-    save_name_critic = f"Critic_{CktGraph().__class__.__name__}_{date}_noise={noise_type}_reward={best_reward:2f}_{GNN().__class__.__name__}_rew_eng={rew_eng}.pth"
+    save_name_critic = f"Critic_{CktGraph().__class__.__name__}_{date}_noise={noise_type}_reward={best_reward:.2f}_{GNN().__class__.__name__}_rew_eng={rew_eng}.pth"
 
     torch.save(model_weight_actor, PWD + "/saved_weights/" + save_name_actor)
     torch.save(model_weight_critic, PWD + "/saved_weights/" + save_name_critic)

python/main2.py

@@ -96,7 +96,7 @@
     save_name_actor = f"Actor_{CktGraph().__class__.__name__}_{date}_noise={noise_type}_reward={best_reward:.2f}_{GNN().__class__.__name__}_rew_eng={rew_eng}.pth"
 
     model_weight_critic = agent.critic.state_dict()
-    save_name_critic = f"Critic_{CktGraph().__class__.__name__}_{date}_noise={noise_type}_reward={best_reward:2f}_{GNN().__class__.__name__}_rew_eng={rew_eng}.pth"
+    save_name_critic = f"Critic_{CktGraph().__class__.__name__}_{date}_noise={noise_type}_reward={best_reward:.2f}_{GNN().__class__.__name__}_rew_eng={rew_eng}.pth"
 
     torch.save(model_weight_actor, PWD + "/saved_weights/" + save_name_actor)
     torch.save(model_weight_critic, PWD + "/saved_weights/" + save_name_critic)

python/models.py

@@ -271,10 +271,10 @@ def __init__(self, CktGraph):
 
             self.in_channels = self.num_node_features
             self.out_channels = self.action_dim
-            self.conv1 = Linear(self.in_channels, 32)
-            self.conv2 = Linear(32, 32)
-            self.conv3 = Linear(32, 16)
-            self.conv4 = Linear(16, 16)
+            self.mlp1 = Linear(self.in_channels, 32)
+            self.mlp2 = Linear(32, 32)
+            self.mlp3 = Linear(32, 16)
+            self.mlp4 = Linear(16, 16)
             self.lin1 = LazyLinear(self.out_channels)
 
         def forward(self, state):
@@ -287,10 +287,10 @@ def forward(self, state):
             actions = torch.tensor(()).to(device)
             for i in range(batch_size):
                 x = state[i]
-                x = F.relu(self.conv1(x))
-                x = F.relu(self.conv2(x))
-                x = F.relu(self.conv3(x))
-                x = F.relu(self.conv4(x))
+                x = F.relu(self.mlp1(x))
+                x = F.relu(self.mlp2(x))
+                x = F.relu(self.mlp3(x))
+                x = F.relu(self.mlp4(x))
                 x = self.lin1(torch.flatten(x))
                 x = torch.tanh(x).reshape(1, -1)
                 actions = torch.cat((actions, x), axis=0)
@@ -308,10 +308,10 @@ def __init__(self, CktGraph):
 
             self.in_channels = self.num_node_features + self.action_dim
             self.out_channels = 1
-            self.conv1 = Linear(self.in_channels, 32)
-            self.conv2 = Linear(32, 32)
-            self.conv3 = Linear(32, 16)
-            self.conv4 = Linear(16, 16)
+            self.mlp1 = Linear(self.in_channels, 32)
+            self.mlp2 = Linear(32, 32)
+            self.mlp3 = Linear(32, 16)
+            self.mlp4 = Linear(16, 16)
             self.lin1 = LazyLinear(self.out_channels)
 
         def forward(self, state, action):
@@ -325,11 +325,11 @@ def forward(self, state, action):
             values = torch.tensor(()).to(device)
             for i in range(batch_size):
                 x = data[i]
-                x = F.relu(self.conv1(x))
-                x = F.relu(self.conv2(x))
-                x = F.relu(self.conv3(x))
-                x = F.relu(self.conv4(x))
+                x = F.relu(self.mlp1(x))
+                x = F.relu(self.mlp2(x))
+                x = F.relu(self.mlp3(x))
+                x = F.relu(self.mlp4(x))
                 x = self.lin1(torch.flatten(x)).reshape(1, -1)
                 values = torch.cat((values, x), axis=0)
 
-            return values
+            return values