diff --git a/targets/TARGET_NUVOTON/nu_timer.c b/targets/TARGET_NUVOTON/nu_timer.c
index ef3d3489ca..a58e7b2f5c 100644
--- a/targets/TARGET_NUVOTON/nu_timer.c
+++ b/targets/TARGET_NUVOTON/nu_timer.c
@@ -65,3 +65,65 @@ void nu_busy_wait_us(uint32_t us)
         prev = cur;
     }
 }
+
+/* Delay 4 cycles per round by hand-counting instruction cycles
+ *
+ * The delay function here is implemented by just hand-counting instruction cycles rather than preferred
+ * H/W timer since it is to use in cases where H/W timer is not available. Usually, it can delay at least
+ * 4-cycles per round.
+ *
+ * In modern pipeline core, plus flash performance and other factors, we cannot rely accurately on hand-
+ * counting instruction cycles for expected delay cycles.
+ */
+#if defined(__CC_ARM)
+MBED_NOINLINE
+__asm void nu_delay_cycle_x4(uint32_t rounds)
+{
+// AStyle should not format inline assembly
+// *INDENT-OFF*
+1
+#if !defined(__CORTEX_M0)
+    NOP                         // 1 cycle
+#endif
+    SUBS a1, a1, #1             // 1 cycle
+    BCS  %BT1                   // 3 cycles(M0)/2 cycles(non-M0)
+    BX   lr
+// *INDENT-ON*
+}
+#elif defined (__ICCARM__)
+MBED_NOINLINE
+void nu_delay_cycle_x4(uint32_t rounds)
+{
+    __asm volatile(
+        "loop: \n"
+#if !defined(__CORTEX_M0)
+        " NOP \n"               // 1 cycle
+#endif
+        " SUBS %0, %0, #1 \n"   // 1 cycle
+        " BCS.n  loop\n"        // 3 cycles(M0)/2 cycles(non-M0)
+        : "+r"(rounds)
+        :
+        : "cc"
+    );
+}
+#elif  defined ( __GNUC__ ) ||  (defined(__ARMCC_VERSION) && (__ARMCC_VERSION >= 6010050))
+MBED_NOINLINE
+void nu_delay_cycle_x4(uint32_t rounds)
+{
+    __asm__ volatile(
+        "%=:\n\t"
+#if !defined(__CORTEX_M0)
+        "NOP\n\t"               // 1 cycle
+#endif
+#if defined(__thumb__) && !defined(__thumb2__) && !defined(__ARMCC_VERSION)
+        "SUB  %0, #1\n\t"       // 1 cycle
+#else
+        "SUBS %0, %0, #1\n\t"   // 1 cycle
+#endif
+        "BCS  %=b\n\t"          // 3 cycles(M0)/2 cycles(non-M0)
+        : "+l"(rounds)
+        :
+        : "cc"
+    );
+}
+#endif
diff --git a/targets/TARGET_NUVOTON/nu_timer.h b/targets/TARGET_NUVOTON/nu_timer.h
index 4b9f359e2e..f3d4d867bb 100644
--- a/targets/TARGET_NUVOTON/nu_timer.h
+++ b/targets/TARGET_NUVOTON/nu_timer.h
@@ -74,6 +74,9 @@ void nu_countdown_free(struct nu_countdown_ctx_s *ctx);
  */
 void nu_busy_wait_us(uint32_t us);
 
+/* Delay 4 cycles per round by hand-counting instruction cycles */
+void nu_delay_cycle_x4(uint32_t rounds);  
+
 #ifdef __cplusplus
 }
 #endif