Rebuilding the Vacuum Fluorescent Display

Back in April 2016 I posted details on how a Vacuum Fluorescent Display (VFD) works and a method of driving it from an Arduino type microcontroller using transistors, while this worked well it made for a rather bulky clock. I have now rebuilt the display using proper VFD driver chips, added a rotary switch with LED’s, NeoPixel capability and used a WeMos D1 Mini microcontroller to give me internet access over WiFi so the time and date can be set using the Network Time Protocol (NTP).

VFD Clock

This rebuild is much more practical than previously, it is now about 3cm deep and comprises of two boards back to back with pin header plugs/sockets as the interconnects. I have been running the clock for over a year without any problems, apart from having to switch the thing off and back on when Daylight Savings Time changes in the spring and autumn, but that’s a software (lazy programmer) problem.

Power Supply

Three different voltages are needed to operate this clock, 16 Volts for the VFD display anodes, 5 Volts for the microcontroller and NeoPixels and 3.3 Volts for the VFD cathode and other hardware.

Power layout for the VFD clock

From the 12 volt source, I split the power to a XL6009 boost module to supply 16V to the driver IC’s, the 5V power module can supply up to 5A which is enough for around 2 meters of NeoPixels. The 3.3V supply is created using a BA033T fixed voltage regulator running from the 5V module, I have used this to power the VFD cathode, the rotary encoder and VFD driver IC’s as this is the microcontrollers operating voltage.

Back of the VFD Clock

Driving the Display

To drive the anodes on the VFD display I used three MAX6920 Tube Drivers, these have a four wire serial interface and can switch up to 76V on the display.

MAX6920 VFD Driver

The physical connections from the driver IC’s to the VFD anodes (grids and elements, see the previous article for definition) are those that are most convenient for the PCB layout without having to use links or through-hole connections as the order of the pins for outputting the display can be mapped later in software.

MAX6920 Anode Connections on the PCB

In this case, the first anode pin on the display is on the right, but the serial data starts with the first MAX6920 chip on the left. I will explain in more detail in the software section below.

The Cathode is run from the 3.3 Volt supply, it has a 3.3V Zenner diode across it to protect against meltdown in case of over-voltage, the cathode is made from of strands of thin tungsten wires and shows as a short when testing in beep mode on the multimeter.

Zenner Diode for the VFD Cathode

Rotary Switch

I have used the SparkFun RGB Illuminated Rotary Encoder, as well as the encoding it also includes three LED’s and a push button switch. As the WeMos only has limited number of pins available, I used an MCP23008 I2C I/O Expander. I have covered these encoders in a previous post and have used a similar method here.

Rotary Encoder on the I2C Bus

In this project it turned out I don’t have much use for the rotary switch, the same result can be achieved with a push button and two LED’s, it goes red when setting the time over WiFi, goes green when the time has finished setting but is off when running normally, the push button displays the date when pushed. I expect the encoder could be used for setting an alarm should I ever have need to add one.


The complete source code I am currently running in the clock can be found on my github repository, it has been written for the Arduino platform and can be considered unfinished and a bit of a mess. In almost all cases this clock will be a custom build with all the scavenged VFD displays being different I’ll be concentrating on how the display is being driven. Note that the software uses Latch and the MAX6920 driver chip uses Load.

To start, here are the physical connections for my VFD display, the display connections are shown looking from the back with pin one on the right, on the anode there are eleven grids and twenty-one elements, each grid illuminates a block of chosen elements. There are three MAX6920 driver chips, the first is on the left data is passed along the serial data line DIN/DOUT to the right:

Connecting the MAX6920 drivers to the display

While I’m not going to worry too much about how the data is sent to the display as this is being handled by the ShiftOutX library it may be useful to have an understanding of the byte order sent to the driver chips.

Breaking out the logic analyser we can see what is going on, in this first image below we can see that between the two markers one and two there are eleven data peaks to match the number of grids on the display the data is processed as a padded eleven character string: ‘    212848 ‘ the first four contain nothing but the bit selecting grid to be used the next six are busy as they contain the time, and the last is blank. There also appear to be two blank clock cycles after these, this may be a programming error on my part.

Data being sent to the display

the string is iterated so that each grid gets its own character with a very short delay between each this causes the persistence of vision effect and the time appears on the display.

In the next screen capture I have zoomed into more detail, it is showing the fifth grid and is displaying the number two. You can see that there are thirty-two clock cycles sent to the drivers while LOAD is high, this causes the driver chips to direct the data to the outputs. I am not entirely sure why the clock is being run while LOAD is low, I think it may be clearing the buffer before sending the required information.

Data being sent to a grid

You can see that 32 bits are sent to the display, but there are 36 pins available on the three MAX6920 drivers, you can also see that the outputs are arranged out of order. How do the drivers know which lines to switch for the display?

First off, an explanation of how the data is constructed then sent to the display. For this example I am sending the number five to the third grid, the data is constructed through the manipulation of a character array in three stages:

1. blank char array: 00000000000000000000000000000000
2. set the third grid: 00000000000000000000000000000100
3. set the number: 00111000011000011100000000000100

Because of the way the bits are ordered when sent to the drivers I have a lookup array VFDlookup to make life easier, in this case grid 3 can be found at position 29 in the VFDdata character array. I start with a blank character array of 32 bits VFDblank and copy this to VFDdata, I then use VFDlookup to find the position of the bit to set: VFDlookup[3] = 29. For the elements the encoding for each character is set in VFDdigits the code for number 5 is 1110000110000111 the code was worked out by experimentation, seeing which element lit up after I had completed the hardware, I have ignored the extra symbols used for the DVD player so I only need to set pins 15-30. Once complete the character array is converted to a unsigned long integer and sent LSBFIRST – Least Significant Bit First via ShiftOut_32.

While that explains the elements encoding to display the correct digit, the grid selection however is a different kettle of fish. I built this clock over a year ago and have forgotten how it works. Grids 1 and 2 are connected to the output pins 1 and 0 on the third driver, pins 2-5 are then skipped, and the rest are connected to the grids, with three more being taken from chip 2. I can’t work out how it knows to skip those pins, or that grid 2 is connected to pin 1. But it is an awesome clock though.

Links and Sources

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.