diff --git a/include/scalehls/Dialect/HLSCpp/Ops.td b/include/scalehls/Dialect/HLSCpp/Ops.td
index 8048854..a35350b 100644
--- a/include/scalehls/Dialect/HLSCpp/Ops.td
+++ b/include/scalehls/Dialect/HLSCpp/Ops.td
@@ -37,7 +37,7 @@ def StreamReadOp : HLSCppOp<"stream.read"> {
 def StreamWriteOp : HLSCppOp<"stream.write"> {
   let summary = "Stream channel write operation";
 
-  let arguments = (ins StreamOf<[AnyType]>:$channel, Optional<AnyType>:$value);
+  let arguments = (ins StreamOf<[AnyType]>:$channel, AnyType:$value);
 }
 
 def StreamBufferOp : HLSCppOp<"stream.buffer"> {
diff --git a/include/scalehls/Dialect/HLSCpp/Visitor.h b/include/scalehls/Dialect/HLSCpp/Visitor.h
index 4d95068..5dc905c 100644
--- a/include/scalehls/Dialect/HLSCpp/Visitor.h
+++ b/include/scalehls/Dialect/HLSCpp/Visitor.h
@@ -52,8 +52,9 @@ public:
             memref::ExpandShapeOp, memref::ReinterpretCastOp,
             bufferization::ToMemrefOp, bufferization::ToTensorOp,
 
-            // HLSCpp primitive operations.
-            PrimMulOp, PrimCastOp, BufferOp,
+            // HLSCpp dialect operations.
+            StreamChannelOp, StreamReadOp, StreamWriteOp, PrimMulOp, PrimCastOp,
+            BufferOp,
 
             // Control flow operations.
             func::CallOp, func::ReturnOp,
@@ -145,7 +146,10 @@ public:
   HANDLE(bufferization::ToMemrefOp);
   HANDLE(bufferization::ToTensorOp);
 
-  // HLSCpp primitive operations.
+  // HLSCpp dialect operations.
+  HANDLE(StreamChannelOp);
+  HANDLE(StreamReadOp);
+  HANDLE(StreamWriteOp);
   HANDLE(PrimMulOp);
   HANDLE(PrimCastOp);
   HANDLE(BufferOp);
diff --git a/lib/Transforms/Graph/CreateTokenFlow.cpp b/lib/Transforms/Graph/CreateTokenFlow.cpp
index a9407d8..1482a2b 100644
--- a/lib/Transforms/Graph/CreateTokenFlow.cpp
+++ b/lib/Transforms/Graph/CreateTokenFlow.cpp
@@ -51,7 +51,9 @@ struct CreateTokenFlow : public CreateTokenFlowBase<CreateTokenFlow> {
         // Create a new token stream channel.
         builder.setInsertionPoint(returnOp);
         auto channel = builder.create<hlscpp::StreamChannelOp>(loc, tokenType);
-        builder.create<hlscpp::StreamWriteOp>(loc, channel, Value());
+        auto tokenValue =
+            builder.create<arith::ConstantOp>(loc, builder.getBoolAttr(false));
+        builder.create<hlscpp::StreamWriteOp>(loc, channel, tokenValue);
 
         // Collect the users of the stream channel.
         for (auto user : result.getUsers()) {
diff --git a/lib/Transforms/Graph/HoistStreamChannel.cpp b/lib/Transforms/Graph/HoistStreamChannel.cpp
index 8db1887..554a3de 100644
--- a/lib/Transforms/Graph/HoistStreamChannel.cpp
+++ b/lib/Transforms/Graph/HoistStreamChannel.cpp
@@ -129,7 +129,10 @@ struct LowerStreamBufferOpRewritePattern
         buffer, buffer.getType());
 
     rewriter.setInsertionPoint(block->getTerminator());
-    rewriter.create<hlscpp::StreamWriteOp>(loc, channel, Value());
+    auto tokenType = channel.getType().cast<StreamType>().getElementType();
+    auto tokenValue = rewriter.create<arith::ConstantOp>(
+        loc, rewriter.getZeroAttr(tokenType));
+    rewriter.create<hlscpp::StreamWriteOp>(loc, channel, tokenValue);
     return success();
   }
 };
diff --git a/lib/Translation/EmitHLSCpp.cpp b/lib/Translation/EmitHLSCpp.cpp
index 8fb9914..278b1da 100644
--- a/lib/Translation/EmitHLSCpp.cpp
+++ b/lib/Translation/EmitHLSCpp.cpp
@@ -26,6 +26,8 @@ static SmallString<16> getTypeName(Value val) {
   auto valType = val.getType();
   if (auto arrayType = val.getType().dyn_cast<ShapedType>())
     valType = arrayType.getElementType();
+  else if (auto streamType = val.getType().dyn_cast<StreamType>())
+    valType = streamType.getElementType();
 
   // Handle float types.
   if (valType.isa<Float32Type>())
@@ -224,7 +226,10 @@ public:
   void emitTensorToMemref(bufferization::ToMemrefOp op);
   void emitMemrefToTensor(bufferization::ToTensorOp op);
 
-  /// HLSCpp primitive operation emitters.
+  /// HLSCpp dialect operation emitters.
+  void emitStreamChannel(StreamChannelOp op);
+  void emitStreamRead(StreamReadOp op);
+  void emitStreamWrite(StreamWriteOp op);
   void emitPrimMul(PrimMulOp op);
   template <typename AssignOpType> void emitAssign(AssignOpType op);
 
@@ -249,7 +254,8 @@ private:
   SmallVector<SmallString<8>, 4> getTransferIndices(TransferOpType op);
 
   /// C++ component emitters.
-  void emitValue(Value val, unsigned rank = 0, bool isPtr = false);
+  void emitValue(Value val, unsigned rank = 0, bool isPtr = false,
+                 bool isRef = false);
   void emitArrayDecl(Value array);
   unsigned emitNestedLoopHeader(Value val);
   void emitNestedLoopFooter(unsigned rank);
@@ -422,7 +428,13 @@ public:
   bool visitOp(bufferization::ToTensorOp op) {
     return emitter.emitMemrefToTensor(op), true;
   }
-  /// HLSCpp primitive operations.
+
+  /// HLSCpp dialect operations.
+  bool visitOp(StreamChannelOp op) {
+    return emitter.emitStreamChannel(op), true;
+  }
+  bool visitOp(StreamReadOp op) { return emitter.emitStreamRead(op), true; }
+  bool visitOp(StreamWriteOp op) { return emitter.emitStreamWrite(op), true; }
   bool visitOp(PrimMulOp op) { return emitter.emitPrimMul(op), true; }
   bool visitOp(PrimCastOp op) { return emitter.emitAssign(op), true; }
   bool visitOp(BufferOp op) { return emitter.emitAssign(op), true; }
@@ -572,8 +584,7 @@ bool ExprVisitor::visitOp(arith::CmpIOp op) {
 
 /// SCF statement emitters.
 void ModuleEmitter::emitScfFor(scf::ForOp op) {
-  indent();
-  os << "for (";
+  indent() << "for (";
   auto iterVar = op.getInductionVar();
 
   // Emit lower bound.
@@ -601,8 +612,7 @@ void ModuleEmitter::emitScfFor(scf::ForOp op) {
   emitBlock(*op.getBody());
   reduceIndent();
 
-  indent();
-  os << "}\n";
+  indent() << "}\n";
 }
 
 void ModuleEmitter::emitScfIf(scf::IfOp op) {
@@ -619,8 +629,7 @@ void ModuleEmitter::emitScfIf(scf::IfOp op) {
     }
   }
 
-  indent();
-  os << "if (";
+  indent() << "if (";
   emitValue(op.getCondition());
   os << ") {";
   emitInfoAndNewLine(op);
@@ -630,15 +639,13 @@ void ModuleEmitter::emitScfIf(scf::IfOp op) {
   reduceIndent();
 
   if (!op.getElseRegion().empty()) {
-    indent();
-    os << "} else {\n";
+    indent() << "} else {\n";
     addIndent();
     emitBlock(op.getElseRegion().front());
     reduceIndent();
   }
 
-  indent();
-  os << "}\n";
+  indent() << "}\n";
 }
 
 void ModuleEmitter::emitScfYield(scf::YieldOp op) {
@@ -664,8 +671,7 @@ void ModuleEmitter::emitScfYield(scf::YieldOp op) {
 
 /// Affine statement emitters.
 void ModuleEmitter::emitAffineFor(AffineForOp op) {
-  indent();
-  os << "for (";
+  indent() << "for (";
   auto iterVar = op.getInductionVar();
 
   // Emit lower bound.
@@ -719,8 +725,7 @@ void ModuleEmitter::emitAffineFor(AffineForOp op) {
   emitBlock(*op.getBody());
   reduceIndent();
 
-  indent();
-  os << "}\n";
+  indent() << "}\n";
 }
 
 void ModuleEmitter::emitAffineIf(AffineIfOp op) {
@@ -737,8 +742,7 @@ void ModuleEmitter::emitAffineIf(AffineIfOp op) {
     }
   }
 
-  indent();
-  os << "if (";
+  indent() << "if (";
   auto constrSet = op.getIntegerSet();
   AffineExprEmitter constrEmitter(state, constrSet.getNumDims(),
                                   op.getOperands());
@@ -763,15 +767,13 @@ void ModuleEmitter::emitAffineIf(AffineIfOp op) {
   reduceIndent();
 
   if (op.hasElse()) {
-    indent();
-    os << "} else {\n";
+    indent() << "} else {\n";
     addIndent();
     emitBlock(*op.getElseBlock());
     reduceIndent();
   }
 
-  indent();
-  os << "}\n";
+  indent() << "}\n";
 }
 
 void ModuleEmitter::emitAffineParallel(AffineParallelOp op) {
@@ -790,8 +792,7 @@ void ModuleEmitter::emitAffineParallel(AffineParallelOp op) {
 
   auto steps = getIntArrayAttrValue(op, op.getStepsAttrName());
   for (unsigned i = 0, e = op.getNumDims(); i < e; ++i) {
-    indent();
-    os << "for (";
+    indent() << "for (";
     auto iterVar = op.getBody()->getArgument(i);
 
     // Emit lower bound.
@@ -825,8 +826,7 @@ void ModuleEmitter::emitAffineParallel(AffineParallelOp op) {
   for (unsigned i = 0, e = op.getNumDims(); i < e; ++i) {
     reduceIndent();
 
-    indent();
-    os << "}\n";
+    indent() << "}\n";
   }
 }
 
@@ -918,8 +918,7 @@ void ModuleEmitter::emitAffineYield(AffineYieldOp op) {
       emitNestedLoopFooter(rank);
     }
   } else if (auto parentOp = dyn_cast<AffineParallelOp>(op->getParentOp())) {
-    indent();
-    os << "if (";
+    indent() << "if (";
     unsigned ivIdx = 0;
     for (auto iv : parentOp.getBody()->getArguments()) {
       emitValue(iv);
@@ -945,8 +944,7 @@ void ModuleEmitter::emitAffineYield(AffineYieldOp op) {
     }
     reduceIndent();
 
-    indent();
-    os << "} else {\n";
+    indent() << "} else {\n";
 
     // Otherwise, generated values will be accumulated/reduced to the
     // current results with corresponding arith::AtomicRMWKind operations.
@@ -1006,8 +1004,7 @@ void ModuleEmitter::emitAffineYield(AffineYieldOp op) {
     }
     reduceIndent();
 
-    indent();
-    os << "}\n";
+    indent() << "}\n";
   }
 }
 
@@ -1212,8 +1209,7 @@ template <typename OpType> void ModuleEmitter::emitReshape(OpType op) {
   assert(!isDeclared(array) && "has been declared before.");
 
   auto arrayType = array.getType().template cast<ShapedType>();
-  indent();
-  os << getTypeName(array) << " (*";
+  indent() << getTypeName(array) << " (*";
 
   // Add the new value to nameTable and emit its name.
   os << addName(array, false);
@@ -1267,7 +1263,35 @@ void ModuleEmitter::emitMemrefToTensor(bufferization::ToTensorOp op) {
   }
 }
 
-/// HLSCpp primitive operation emitters.
+/// HLSCpp dialect operation emitters.
+void ModuleEmitter::emitStreamChannel(StreamChannelOp op) {
+  indent();
+  emitValue(op.channel());
+  os << ";";
+  emitInfoAndNewLine(op);
+}
+
+void ModuleEmitter::emitStreamRead(StreamReadOp op) {
+  indent();
+  if (op.result()) {
+    emitValue(op.result());
+    os << " = ";
+  }
+  emitValue(op.channel());
+  os << ".read(";
+  os << ");";
+  emitInfoAndNewLine(op);
+}
+
+void ModuleEmitter::emitStreamWrite(StreamWriteOp op) {
+  indent();
+  emitValue(op.channel());
+  os << ".write(";
+  emitValue(op.value());
+  os << ");";
+  emitInfoAndNewLine(op);
+}
+
 void ModuleEmitter::emitPrimMul(PrimMulOp op) {
   if (op.isPackMul()) {
     // Declare the result C array.
@@ -1334,8 +1358,7 @@ void ModuleEmitter::emitCall(func::CallOp op) {
   }
 
   // Emit the function call.
-  indent();
-  os << op.getCallee() << "(";
+  indent() << op.getCallee() << "(";
 
   // Handle input arguments.
   unsigned argIdx = 0;
@@ -1470,7 +1493,8 @@ void ModuleEmitter::emitConstant(arith::ConstantOp op) {
 }
 
 /// C++ component emitters.
-void ModuleEmitter::emitValue(Value val, unsigned rank, bool isPtr) {
+void ModuleEmitter::emitValue(Value val, unsigned rank, bool isPtr,
+                              bool isRef) {
   assert(!(rank && isPtr) && "should be either an array or a pointer.");
 
   // Value has been declared before or is a constant number.
@@ -1481,7 +1505,13 @@ void ModuleEmitter::emitValue(Value val, unsigned rank, bool isPtr) {
     return;
   }
 
-  os << getTypeName(val) << " ";
+  if (val.getType().isa<StreamType>())
+    os << "hls::stream<" << getTypeName(val) << "> ";
+  else
+    os << getTypeName(val) << " ";
+
+  if (isRef)
+    os << "&";
 
   // Add the new value to nameTable and emit its name.
   os << addName(val, isPtr);
@@ -1527,8 +1557,7 @@ unsigned ModuleEmitter::emitNestedLoopHeader(Value val) {
     // Create nested loop.
     unsigned dimIdx = 0;
     for (auto &shape : type.getShape()) {
-      indent();
-      os << "for (int iv" << dimIdx << " = 0; ";
+      indent() << "for (int iv" << dimIdx << " = 0; ";
       os << "iv" << dimIdx << " < " << shape << "; ";
       os << "++iv" << dimIdx++ << ") {\n";
 
@@ -1548,8 +1577,7 @@ void ModuleEmitter::emitNestedLoopFooter(unsigned rank) {
   for (unsigned i = 0; i < rank; ++i) {
     reduceIndent();
 
-    indent();
-    os << "}\n";
+    indent() << "}\n";
   }
 }
 
@@ -1589,14 +1617,10 @@ void ModuleEmitter::emitLoopDirectives(Operation *op) {
   if (!loopDirect)
     return;
 
-  if (loopDirect.getPipeline()) {
-    indent();
-    os << "#pragma HLS pipeline II=" << loopDirect.getTargetII() << "\n";
-
-  } else if (loopDirect.getDataflow()) {
-    indent();
-    os << "#pragma HLS dataflow\n";
-  }
+  if (loopDirect.getPipeline())
+    indent() << "#pragma HLS pipeline II=" << loopDirect.getTargetII() << "\n";
+  else if (loopDirect.getDataflow())
+    indent() << "#pragma HLS dataflow\n";
 }
 
 void ModuleEmitter::emitArrayDirectives(Value memref) {
@@ -1612,8 +1636,7 @@ void ModuleEmitter::emitArrayDirectives(Value memref) {
     if (factors[dim] != 1) {
       emitPragmaFlag = true;
 
-      indent();
-      os << "#pragma HLS array_partition";
+      indent() << "#pragma HLS array_partition";
       os << " variable=";
       emitValue(memref);
 
@@ -1634,8 +1657,7 @@ void ModuleEmitter::emitArrayDirectives(Value memref) {
   if (kind != MemoryKind::DRAM && !isFullyPartitioned(type)) {
     emitPragmaFlag = true;
 
-    indent();
-    os << "#pragma HLS resource";
+    indent() << "#pragma HLS resource";
     os << " variable=";
     emitValue(memref);
 
@@ -1651,6 +1673,16 @@ void ModuleEmitter::emitArrayDirectives(Value memref) {
     os << "\n";
   }
 
+  // Emit DRAM variable as stable.
+  if (kind == MemoryKind::DRAM) {
+    emitPragmaFlag = true;
+
+    indent() << "#pragma HLS stable";
+    os << " variable=";
+    emitValue(memref);
+    os << "\n";
+  }
+
   // Emit an empty line.
   if (emitPragmaFlag)
     os << "\n";
@@ -1660,8 +1692,7 @@ void ModuleEmitter::emitFunctionDirectives(FuncOp func,
                                            ArrayRef<Value> portList) {
   // Only top function should emit interface pragmas.
   if (hasTopFuncAttr(func)) {
-    indent();
-    os << "#pragma HLS interface s_axilite port=return bundle=ctrl\n";
+    indent() << "#pragma HLS interface s_axilite port=return bundle=ctrl\n";
 
     for (auto &port : portList) {
       // Array ports and scalar ports are handled separately. Here, we only
@@ -1669,8 +1700,7 @@ void ModuleEmitter::emitFunctionDirectives(FuncOp func,
       if (auto memrefType = port.getType().dyn_cast<MemRefType>()) {
         // Only emit interface pragma when the array is not fully partitioned.
         if (!isFullyPartitioned(memrefType)) {
-          indent();
-          os << "#pragma HLS interface";
+          indent() << "#pragma HLS interface";
           // For now, we set the offset of all m_axi interfaces as slave.
           if (MemoryKind(memrefType.getMemorySpaceAsInt()) ==
               MemoryKind::DRAM) {
@@ -1684,8 +1714,7 @@ void ModuleEmitter::emitFunctionDirectives(FuncOp func,
           os << "\n";
         }
       } else {
-        indent();
-        os << "#pragma HLS interface s_axilite";
+        indent() << "#pragma HLS interface s_axilite";
         os << " port=";
 
         // TODO: This is a temporary solution.
@@ -1711,14 +1740,13 @@ void ModuleEmitter::emitFunctionDirectives(FuncOp func,
     return;
 
   if (funcDirect.getPipeline()) {
-    indent();
-    os << "#pragma HLS pipeline II=" << funcDirect.getTargetInterval() << "\n";
+    indent() << "#pragma HLS pipeline II=" << funcDirect.getTargetInterval()
+             << "\n";
 
     // An empty line.
     os << "\n";
   } else if (funcDirect.getDataflow()) {
-    indent();
-    os << "#pragma HLS dataflow\n";
+    indent() << "#pragma HLS dataflow\n";
 
     // An empty line.
     os << "\n";
@@ -1758,6 +1786,8 @@ void ModuleEmitter::emitFunction(FuncOp func) {
     indent();
     if (arg.getType().isa<ShapedType>())
       emitArrayDecl(arg);
+    else if (arg.getType().isa<StreamType>())
+      emitValue(arg, /*rank=*/0, /*isPtr=*/false, /*isRef=*/true);
     else
       emitValue(arg);