diff --git a/apa102/apa102.go b/apa102/apa102.go
index 4a8ab9a..98253bf 100644
--- a/apa102/apa102.go
+++ b/apa102/apa102.go
@@ -27,44 +27,46 @@ var startFrame = []byte{0x00, 0x00, 0x00, 0x00}
 type Device struct {
 	bus   drivers.SPI
 	Order int
+	buf   [4]byte
 }
 
 // New returns a new APA102 driver. Pass in a fully configured SPI bus.
-func New(b drivers.SPI) Device {
-	return Device{bus: b, Order: BGR}
+func New(b drivers.SPI) *Device {
+	return &Device{bus: b, Order: BGR}
 }
 
 // NewSoftwareSPI returns a new APA102 driver that will use a software based
 // implementation of the SPI protocol.
-func NewSoftwareSPI(sckPin, sdoPin machine.Pin, delay uint32) Device {
+func NewSoftwareSPI(sckPin, sdoPin machine.Pin, delay uint32) *Device {
 	return New(&bbSPI{SCK: sckPin, SDO: sdoPin, Delay: delay})
 }
 
 // WriteColors writes the given RGBA color slice out using the APA102 protocol.
 // The A value (Alpha channel) is used for brightness, set to 0xff (255) for maximum.
-func (d Device) WriteColors(cs []color.RGBA) (n int, err error) {
+func (d *Device) WriteColors(cs []color.RGBA) (n int, err error) {
 	d.startFrame()
 
 	// write data
 	for _, c := range cs {
 		// brightness is scaled to 5 bit value
-		d.bus.Transfer(0xe0 | (c.A >> 3))
+		d.buf[0] = 0xe0 | (c.A >> 3)
 
 		// set the colors
 		switch d.Order {
 		case BRG:
-			d.bus.Transfer(c.B)
-			d.bus.Transfer(c.R)
-			d.bus.Transfer(c.G)
+			d.buf[1] = c.B
+			d.buf[2] = c.R
+			d.buf[3] = c.G
 		case GRB:
-			d.bus.Transfer(c.G)
-			d.bus.Transfer(c.R)
-			d.bus.Transfer(c.B)
+			d.buf[1] = c.G
+			d.buf[2] = c.R
+			d.buf[3] = c.B
 		case BGR:
-			d.bus.Transfer(c.B)
-			d.bus.Transfer(c.G)
-			d.bus.Transfer(c.R)
+			d.buf[1] = c.B
+			d.buf[2] = c.G
+			d.buf[3] = c.R
 		}
+		d.bus.Tx(d.buf[:], nil)
 	}
 
 	d.endFrame(len(cs))
@@ -73,7 +75,7 @@ func (d Device) WriteColors(cs []color.RGBA) (n int, err error) {
 }
 
 // Write the raw bytes using the APA102 protocol.
-func (d Device) Write(buf []byte) (n int, err error) {
+func (d *Device) Write(buf []byte) (n int, err error) {
 	d.startFrame()
 	d.bus.Tx(buf, nil)
 	d.endFrame(len(buf) / 4)
@@ -82,14 +84,14 @@ func (d Device) Write(buf []byte) (n int, err error) {
 }
 
 // startFrame sends the start bytes for a strand of LEDs.
-func (d Device) startFrame() {
+func (d *Device) startFrame() {
 	d.bus.Tx(startFrame, nil)
 }
 
 // endFrame sends the end frame marker with one extra bit per LED so
 // long strands of LEDs receive the necessary termination for updates.
 // See https://cpldcpu.wordpress.com/2014/11/30/understanding-the-apa102-superled/
-func (d Device) endFrame(count int) {
+func (d *Device) endFrame(count int) {
 	for i := 0; i < count/16; i++ {
 		d.bus.Transfer(0xff)
 	}
diff --git a/examples/apa102/itsybitsy-m0/main.go b/examples/apa102/itsybitsy-m0/main.go
index 1cfdb70..ac6d8c6 100644
--- a/examples/apa102/itsybitsy-m0/main.go
+++ b/examples/apa102/itsybitsy-m0/main.go
@@ -13,7 +13,7 @@ import (
 )
 
 var (
-	apa apa102.Device
+	apa *apa102.Device
 
 	pwm   = machine.TCC0
 	leds  = make([]color.RGBA, 1)