diff --git a/substrate/client/consensus/pow/README.md b/substrate/client/consensus/pow/README.md
index a335ec367047b9762cc01132da1b39e3d4d3ed0d..8dba30fc5a38e64ada65f068f327143eb40962b3 100644
--- a/substrate/client/consensus/pow/README.md
+++ b/substrate/client/consensus/pow/README.md
@@ -3,7 +3,15 @@ Proof of work consensus for Substrate.
 To use this engine, you can need to have a struct that implements
 `PowAlgorithm`. After that, pass an instance of the struct, along
 with other necessary client references to `import_queue` to setup
-the queue. Use the `start_mine` function for basic CPU mining.
+the queue.
+
+This library also comes with an async mining worker, which can be
+started via the `start_mining_worker` function. It returns a worker
+handle together with a future. The future must be pulled. Through
+the worker handle, you can pull the metadata needed to start the
+mining process via `MiningWorker::metadata`, and then do the actual
+mining on a standalone thread. Finally, when a seal is found, call
+`MiningWorker::submit` to build the block.
 
 The auxiliary storage for PoW engine only stores the total difficulty.
 For other storage requirements for particular PoW algorithm (such as
@@ -13,4 +21,4 @@ for the auxiliary storage. It is also possible to just use the runtime
 as the storage, but it is not recommended as it won't work well with light
 clients.
 
-License: GPL-3.0-or-later WITH Classpath-exception-2.0
\ No newline at end of file
+License: GPL-3.0-or-later WITH Classpath-exception-2.0
diff --git a/substrate/client/consensus/pow/src/lib.rs b/substrate/client/consensus/pow/src/lib.rs
index b12bad7bac22ec4eddb27ee8ddf8f71299ee1d50..17cdae48cdb67d8b8fa4458803be55b64f73a440 100644
--- a/substrate/client/consensus/pow/src/lib.rs
+++ b/substrate/client/consensus/pow/src/lib.rs
@@ -19,14 +19,22 @@
 //! Proof of work consensus for Substrate.
 //!
 //! To use this engine, you can need to have a struct that implements
-//! `PowAlgorithm`. After that, pass an instance of the struct, along
-//! with other necessary client references to `import_queue` to setup
-//! the queue. Use the `start_mine` function for basic CPU mining.
+//! [`PowAlgorithm`]. After that, pass an instance of the struct, along
+//! with other necessary client references to [`import_queue`] to setup
+//! the queue.
+//!
+//! This library also comes with an async mining worker, which can be
+//! started via the [`start_mining_worker`] function. It returns a worker
+//! handle together with a future. The future must be pulled. Through
+//! the worker handle, you can pull the metadata needed to start the
+//! mining process via [`MiningWorker::metadata`], and then do the actual
+//! mining on a standalone thread. Finally, when a seal is found, call
+//! [`MiningWorker::submit`] to build the block.
 //!
 //! The auxiliary storage for PoW engine only stores the total difficulty.
 //! For other storage requirements for particular PoW algorithm (such as
 //! the actual difficulty for each particular blocks), you can take a client
-//! reference in your `PowAlgorithm` implementation, and use a separate prefix
+//! reference in your [`PowAlgorithm`] implementation, and use a separate prefix
 //! for the auxiliary storage. It is also possible to just use the runtime
 //! as the storage, but it is not recommended as it won't work well with light
 //! clients.