diff --git a/src/main/scala/funcunit/BranchControl.scala b/src/main/scala/funcunit/BranchControl.scala
index b89af1b..52d9198 100644
--- a/src/main/scala/funcunit/BranchControl.scala
+++ b/src/main/scala/funcunit/BranchControl.scala
@@ -7,13 +7,6 @@ import pipeline._
 
 object BranchControl extends Enumeration {
   val TYPE_BEQ, TYPE_BNE, TYPE_JAL, TYPE_JR, NR_BR_INST = Value
-//  val TYPE_BEQ = 0
-//  val TYPE_BNE = 1
-
-//  val TYPE_JAL = 2
-//  val TYPE_JR  = 3
-
-//  val NR_BR_INST = 4
 }
 
 class BranchControl extends RawModule{
diff --git a/src/main/scala/funcunit/ForwardPath.scala b/src/main/scala/funcunit/ForwardPath.scala
new file mode 100644
index 0000000..b8a312f
--- /dev/null
+++ b/src/main/scala/funcunit/ForwardPath.scala
@@ -0,0 +1,43 @@
+package funcunit
+
+import chisel3._
+import pipeline.{DsToFPBus, EsToFPBus, MsToFPBus}
+
+class ForwardPath extends Module{
+  val es2fp_bus = IO(Flipped(new EsToFPBus))
+  val ms2fp_bus = IO(Flipped(new MsToFPBus))
+  val ds2fp_bus = IO(Flipped(new DsToFPBus))
+  val rf_read = IO(Flipped(new RFRead))
+
+  val es_rel = Wire(Vec(2, Bool()))
+  val ms_rel = Wire(Vec(2, Bool()))
+  val es_hit = Wire(Vec(2, Bool()))
+  val ms_hit = Wire(Vec(2, Bool()))
+
+  rf_read.raddr := ds2fp_bus.rf_read.raddr
+
+  for (i <- 0 to 1) {
+    es_rel(i) := (ds2fp_bus.rf_read.raddr(i) === es2fp_bus.data.rf_wnum) && es2fp_bus.data.addr_valid && (ds2fp_bus.rf_read.raddr(i) =/= 0.U)
+    es_hit(i) := es2fp_bus.data.data_valid
+
+    ms_rel(i) := (ds2fp_bus.rf_read.raddr(i) === ms2fp_bus.data.rf_wnum) && ms2fp_bus.data.addr_valid && (ds2fp_bus.rf_read.raddr(i) =/= 0.U)
+    ms_hit(i) := ms2fp_bus.data.data_valid
+
+    when (es_rel(i) && es_hit(i)) {
+      ds2fp_bus.rf_read.rdata(i) := es2fp_bus.data.rf_wdata
+    }.elsewhen(ms_rel(i) && ms_hit(i)) {
+      ds2fp_bus.rf_read.rdata(i) := ms2fp_bus.data.rf_wdata
+    }.otherwise {
+      ds2fp_bus.rf_read.rdata(i) := rf_read.rdata(i)
+    }
+
+
+  }
+
+  when (es_rel(0) && !es_hit(0) || es_rel(1) && !es_hit(1)) {
+    ds2fp_bus.stall := true.B
+  }.otherwise {
+    ds2fp_bus.stall := false.B
+  }
+
+}
diff --git a/src/main/scala/mycpu_top.scala b/src/main/scala/mycpu_top.scala
index b4f334f..b7db3c4 100644
--- a/src/main/scala/mycpu_top.scala
+++ b/src/main/scala/mycpu_top.scala
@@ -20,15 +20,16 @@ class mycpu_top extends Module {
     val es = Module(new Execute)
     val ms = Module(new Memory)
     val bc = Module(new BranchControl)
+    val fp = Module(new ForwardPath)
     val rf = Module(new RegFile)
 
     // Pipeline Interface
     fs.fs2ds_bus <> ds.fs2ds_bus
     ds.ds2es_bus <> es.ds2es_bus
-    es.es2ms_bus <> ms.es2msbus
+    es.es2ms_bus <> ms.es2ms_bus
 
     // RegFile Interface
-    ds.rf_read <> rf.read_channel
+    fp.rf_read <> rf.read_channel
     ms.rf_write <> rf.write_channel
 
     // Sram Interface
@@ -47,5 +48,10 @@ class mycpu_top extends Module {
     bc.es2bc_bus <> es.es2bc_bus
     bc.bc2ds_bus <> ds.bc2ds_bus
     bc.bc2fs_bus <> fs.bc2fs_bus
+
+    // Forward Path
+    fp.es2fp_bus <> es.es2fp_bus
+    fp.ms2fp_bus <> ms.ms2fp_bus
+    fp.ds2fp_bus <> ds.data_fetch
   }
 }
diff --git a/src/main/scala/pipeline/Decode.scala b/src/main/scala/pipeline/Decode.scala
index d68405e..c2fc520 100644
--- a/src/main/scala/pipeline/Decode.scala
+++ b/src/main/scala/pipeline/Decode.scala
@@ -14,7 +14,7 @@ class Decode extends Module {
   val ds2es_bus = IO(new DsToEsBus)
   val bc2ds_bus = IO(Flipped(new BcToDsBus))
   val ds2bc_bus = IO(new DsToBcBus)
-  val rf_read = IO(Flipped(new RFRead))
+  val data_fetch = IO(new DsToFPBus)
 
   // Stage Control
   val valid = RegInit(false.B)
@@ -35,7 +35,7 @@ class Decode extends Module {
   // Operand Helper Methods
   def ITypeOp(aluop: Int): Unit = {
     ds2es_bus.data.aluop(aluop) := true.B
-    ds2es_bus.data.src(0) := rf_read.rdata(0)
+    ds2es_bus.data.src(0) := data_fetch.rf_read.rdata(0)
     ds2es_bus.data.src(1) := simm16
     ds2es_bus.data.rf_wen := true.B
     ds2es_bus.data.rf_wnum := rt
@@ -43,14 +43,14 @@ class Decode extends Module {
 
   def ITypeOp_Br(aluop: Int, br_type: Int): Unit = {
     ds2es_bus.data.aluop(aluop) := true.B
-    ds2es_bus.data.src(0) := rf_read.rdata(0)
-    ds2es_bus.data.src(1) := rf_read.rdata(1)
+    ds2es_bus.data.src(0) := data_fetch.rf_read.rdata(0)
+    ds2es_bus.data.src(1) := data_fetch.rf_read.rdata(1)
     ds2es_bus.data.br_offset := simm18
     ds2es_bus.data.br_type(br_type) := true.B
   }
 
   def ItypeOp_Ld(ld_type: Int): Unit = {
-    ds2es_bus.data.src(0) := rf_read.rdata(0)
+    ds2es_bus.data.src(0) := data_fetch.rf_read.rdata(0)
     ds2es_bus.data.src(1) := simm16
     ds2es_bus.data.rf_wen := true.B
     ds2es_bus.data.rf_wnum := rt
@@ -59,18 +59,18 @@ class Decode extends Module {
   }
 
   def ItypeOp_St(st_type: Int): Unit = {
-    ds2es_bus.data.src(0) := rf_read.rdata(0)
+    ds2es_bus.data.src(0) := data_fetch.rf_read.rdata(0)
     ds2es_bus.data.src(1) := simm16
     ds2es_bus.data.req_mem := true.B
     ds2es_bus.data.ld_st_type(st_type) := true.B
     // now br_offset is store data
-    ds2es_bus.data.br_offset := rf_read.rdata(1)
+    ds2es_bus.data.br_offset := data_fetch.rf_read.rdata(1)
   }
 
   def RTypeOp(aluop: Int): Unit = {
     ds2es_bus.data.aluop(aluop) := true.B
-    ds2es_bus.data.src(0) := rf_read.rdata(0)
-    ds2es_bus.data.src(1) := rf_read.rdata(1)
+    ds2es_bus.data.src(0) := data_fetch.rf_read.rdata(0)
+    ds2es_bus.data.src(1) := data_fetch.rf_read.rdata(1)
     ds2es_bus.data.rf_wen := true.B
     ds2es_bus.data.rf_wnum := rd
   }
@@ -78,14 +78,14 @@ class Decode extends Module {
   def RTypeOp_ShiftSa(aluop: Int): Unit = {
     ds2es_bus.data.aluop(aluop) := true.B
     ds2es_bus.data.src(0) := sa
-    ds2es_bus.data.src(1) := rf_read.rdata(1)
+    ds2es_bus.data.src(1) := data_fetch.rf_read.rdata(1)
     ds2es_bus.data.rf_wen := true.B
     ds2es_bus.data.rf_wnum := rd
   }
 
   def RTypeOp_Jr(): Unit = {
     ds2es_bus.data.br_type(TYPE_JR.id) := true.B
-    ds2es_bus.data.br_offset := rf_read.rdata(0)
+    ds2es_bus.data.br_offset := data_fetch.rf_read.rdata(0)
   }
 
   def JTypeOp_Jal(): Unit = {
@@ -125,11 +125,11 @@ class Decode extends Module {
 
   ds2bc_bus.data.pc := data.pc
 
-  rf_read.raddr(0) := rs
-  rf_read.raddr(1) := rt
+  data_fetch.rf_read.raddr(0) := rs
+  data_fetch.rf_read.raddr(1) := rt
 
   /* Stage Control Implement */
-  ds_ready_go := true.B
+  ds_ready_go := !data_fetch.stall
   when (bc2ds_bus.br_taken) {
     valid := false.B
   }.elsewhen(fs2ds_bus.ds_allowin) {
diff --git a/src/main/scala/pipeline/Execute.scala b/src/main/scala/pipeline/Execute.scala
index bfa7967..3aef946 100644
--- a/src/main/scala/pipeline/Execute.scala
+++ b/src/main/scala/pipeline/Execute.scala
@@ -9,6 +9,7 @@ class Execute extends Module{
   val ds2es_bus = IO(Flipped(new DsToEsBus))
   val es2ms_bus = IO(new EsToMsBus)
   val es2bc_bus = IO(new EsToBcBus)
+  val es2fp_bus = IO(new EsToFPBus)
   val data_sram_req = IO(new SramReq)
 
   // Stage Control
@@ -37,6 +38,11 @@ class Execute extends Module{
   es2bc_bus.data.br_type := data.br_type
   es2bc_bus.data.valid := valid
 
+  es2fp_bus.data.rf_wnum := data.rf_wnum
+  es2fp_bus.data.rf_wdata := alu.io.alu_result
+  es2fp_bus.data.addr_valid := valid && data.rf_wen
+  es2fp_bus.data.data_valid := valid && !(data.ld_st_type.asUInt.orR) && data.rf_wen
+
   data_sram_req.en := data.req_mem && valid
   data_sram_req.wen := Mux(data.ld_st_type(1), "b1111".U, 0.U)
   data_sram_req.addr := alu.io.mem_addr_result
diff --git a/src/main/scala/pipeline/Interface.scala b/src/main/scala/pipeline/Interface.scala
index 2ebebab..6132504 100644
--- a/src/main/scala/pipeline/Interface.scala
+++ b/src/main/scala/pipeline/Interface.scala
@@ -85,7 +85,6 @@ class EsToMsBus extends Bundle {
 }
 
 class EsToBcData extends Bundle {
-//  val pc = Output(UInt(32.W))
   val br_offset = Output(UInt(32.W))
   val br_cond = new Bundle() {
     val result = Output(UInt(32.W))
@@ -101,6 +100,15 @@ class EsToBcBus extends Bundle {
   val data = new EsToBcData
 }
 
+class EsToFPBus extends Bundle {
+  val data = new StageToFPData
+}
+
+// memory and forward
+class MsToFPBus extends Bundle {
+  val data = new StageToFPData
+}
+
 // Debug Interface
 class DebugInterface extends Bundle {
   val pc = Output(UInt(32.W))
@@ -120,3 +128,16 @@ class BcToFsBus extends Bundle {
   val br_taken = Output(Bool())
   val br_target = Output(UInt(32.W))
 }
+
+// Forward Path Interface
+class StageToFPData extends Bundle {
+  val rf_wnum = Output(UInt(5.W))
+  val rf_wdata = Output(UInt(32.W))
+  val data_valid = Output(Bool())
+  val addr_valid = Output(Bool())
+}
+
+class DsToFPBus extends Bundle {
+  val rf_read = Flipped(new RFRead)
+  val stall = Input(Bool())
+}
diff --git a/src/main/scala/pipeline/Memory.scala b/src/main/scala/pipeline/Memory.scala
index 4fd2b86..608db1f 100644
--- a/src/main/scala/pipeline/Memory.scala
+++ b/src/main/scala/pipeline/Memory.scala
@@ -7,7 +7,8 @@ import funcunit._
 class Memory extends Module{
   // Interface
   val data_sram_res = IO(new SramRes)
-  val es2msbus = IO(Flipped(new EsToMsBus))
+  val es2ms_bus = IO(Flipped(new EsToMsBus))
+  val ms2fp_bus = IO(new MsToFPBus)
   val rf_write = IO(Flipped(new RFWrite))
   val debug = IO(new DebugInterface)
 
@@ -17,22 +18,29 @@ class Memory extends Module{
   val data = Reg(new EsToMsData)
 
   /* Interface Implement */
-  es2msbus.ms_allowin := !valid || ms_ready_go
+  es2ms_bus.ms_allowin := !valid || ms_ready_go
+
+  ms2fp_bus.data.rf_wnum := rf_write.wnum
+  ms2fp_bus.data.rf_wdata := rf_write.wdata
+  ms2fp_bus.data.addr_valid := valid && data.rf.wen(0)
+  ms2fp_bus.data.data_valid := valid && data.rf.wen(0)
+
   rf_write.wen := data.rf.wen & Fill(4, valid)
   rf_write.wdata := Mux(data.req_mem, data_sram_res.rdata, data.rf.wdata)
   rf_write.wnum := data.rf.wnum
+
   debug.pc := data.pc
   debug.rf := rf_write
 
 
   /* Stage Control Implement */
   ms_ready_go := true.B
-  when (es2msbus.ms_allowin) {
-    valid := es2msbus.es_valid
+  when (es2ms_bus.ms_allowin) {
+    valid := es2ms_bus.es_valid
   }
 
-  when (es2msbus.es_valid && es2msbus.ms_allowin) {
-    data := es2msbus.data
+  when (es2ms_bus.es_valid && es2ms_bus.ms_allowin) {
+    data := es2ms_bus.data
   }
   
 }