diff --git a/itsybitsy_m0_lights/src/lib.rs b/itsybitsy_m0_lights/src/lib.rs
index aba4915..e8ffef8 100644
--- a/itsybitsy_m0_lights/src/lib.rs
+++ b/itsybitsy_m0_lights/src/lib.rs
@@ -1,4 +1,5 @@
 #![no_std]
+#![feature(asm)]
 
 extern crate cortex_m;
 extern crate embedded_hal;
@@ -9,6 +10,7 @@ pub use cortex_m::asm::delay;
 pub use hal::entry;
 
 use core::fmt::Debug;
+use core::sync::atomic::{compiler_fence, Ordering};
 use core::u8;
 pub use embedded_hal::digital::v2::OutputPin;
 use hal::{clock::GenericClockController, pac::Peripherals};
@@ -67,14 +69,42 @@ where
 {
     #[inline]
     pub fn write(&mut self, bit: bool) {
+        // go high
         self.high_out.set_high().unwrap();
-        delay(if bit {
-            ONE_HIGH_CYCLES
-        } else {
-            ZERO_HIGH_CYCLES
-        });
+
+        unsafe {
+            compiler_fence(Ordering::SeqCst);
+            asm!(
+                // ? nops
+                "nop;", "nop;", "nop;", "nop;", "nop;", "nop;", "nop;", "nop;", "nop;",
+            );
+            compiler_fence(Ordering::SeqCst);
+        }
+
+        if bit {
+            unsafe {
+                compiler_fence(Ordering::SeqCst);
+                asm!(
+                    // 7 nops
+                    "nop;", "nop;", "nop;", "nop;", "nop;", "nop;", "nop;", "nop;", "nop;", "nop;",
+                    "nop;", "nop;", "nop;", "nop;",
+                );
+                compiler_fence(Ordering::SeqCst);
+            }
+        }
+
+        // go low
         self.high_out.set_low().unwrap();
-        delay(if bit { ONE_LOW_CYCLES } else { ZERO_LOW_CYCLES });
+
+        unsafe {
+            compiler_fence(Ordering::SeqCst);
+            asm!(
+                // 2 nops
+                "nop;", "nop;", "nop;", "nop;", "nop;", "nop;", "nop;", "nop;", "nop;", "nop;",
+                "nop;", "nop;", "nop;", "nop;", "nop;",
+            );
+            compiler_fence(Ordering::SeqCst);
+        }
     }
 
     #[inline]
@@ -87,7 +117,14 @@ where
         self.write(byte & 0x04 != 0);
         self.write(byte & 0x02 != 0);
         self.write(byte & 0x01 != 0);
-        delay(5);
+        unsafe {
+            compiler_fence(Ordering::SeqCst);
+            asm!(
+                // 5 nops
+                "nop;", "nop;", "nop;", "nop;", "nop;",
+            );
+            compiler_fence(Ordering::SeqCst);
+        }
     }
 
     /// Blink light, probably to indicate a loop is still running