Well, if custom build devices also count, I would present the SLS-Guckies :))
It was a quite unusual terminal system custom build based on telephone keyboards and TV sets, build for a none existing in house solution with no budget:))
By the mid 1970s the mainframe service department for Bavaria of a major manufacturer had already outgrown management by paper records, punch cards, card tables and electro-mechanical clocks (*1). So why not using one of the machines they are familiar with? Except, there was no budget and no way to get one the size needed. Getting a machine wasn't an issue as big, as by the mid 1970 early machines low end were already taken out of service - in this case a 4004/15, about the same class as an equivalent of an IBM 360/20, was found. Together with two 7 MiB Disk drives and the usual (dated) peripherals.
A harder to crack issue was the user side as terminals were still a new thing far from being decommissioned. A new mainframe terminal had a price of a basic VW Bug of the same time, so a clear no-go. For programming this was resolved by using the machines console (*2), but a solution was needed for the user side.
At that point it's important that the head of that department was an engineer and tinkerer by heart (*3) well versed in radio/TV technology. He came up with the idea of using an inexpensive TV and phone based solution:
- A numeric, telephone keyboard based input station.
- A controller
- connected as standard blockmux to mainframe channel,
- forwarding the input to the mainframe
- receivng output from the mainframe
- storing that in TV cards
- which in turn fed modulators connected to coax cables
- Cheap B&W TV sets.
The first part was based of a (back then) brand new touch tone encoder chip which only has been recently finished at the semiconductor plant in Munich - he had gathered a tube during a visit, which might have been the origin of he whole idea. Those chips scanned a 4x4 keyboards, buffered up to 16 key presses (*4) and released them according to touch tone timing rules. Those chips combined with a 12 key keypad were made into the input devices, connected via a two wire telephone line. Well, also with a cranked up clock, so input was send out at typing speed (*5). The controller would listen to all input channels, collect one message at a time and forward it via a device interrupt to the mainframe. All hard wired TTL logic, no microprocessor involved.
For output purpose there were video buffer boards, one per channel. They were as well build without any microprocessor, but a common set of counters to fill them as well one to generate display. They were based 256x1 static RAM devices building a 768 by 6 memory (*6) and one of those brand new monolithic character generators. A bit like what Wozniak did with the Apple II not much later. Display was 24 lines and 32 characters per line at 50 Hz.
Output of those frame buffers was modulated to a TV channel with up to 16 channels per cable. 'Only' 16 as it had to be within reach of cheap analogue TV of that time. An output message from the mainframe started with an address selecting sub frame and video buffer followed by 768 bytes to be outputted.
One sub frame of the controller consisted of (up to) 32 keyboard input channels and 16 frame buffer cards. Sounds strange, but to increase capacity two users (defined per keyboard) would share one output channel. While high performance users had their own, low thruput users wold share one (*7) It was a great way to keep up with the sudden demand :))
Later one double buffer was added, build from two frame buffers, one supplying the characters, the other adding a colour channel. This was used for a big screen TV to give the call center operators a quick status overview for all open cases.
Bottom Line: Anyway, yes, such stuff was build and I don't think we were the only ones to come up with that idea.
We also tried to sell this to management and marketing as a product idea, but it was flat out rejected every time, as they didn't see a market. After all, 'real' customers wanted 'real' terminals - or possible didn't see the sales bonus generated by selling a solution serving 16 users at the cost of a single terminal :))
The whole setup was not just used until the 1980s when terminals became more affordable, but way into the 1990 (I think it was finally switched of in 1997) as many users wanted to keep their fast little independent Guckies in parallel. By that time it had been extended to three sub frames for 48 output channels (well 47 as the colour screen needed two) and 96 keyboards.
To build all of this he drafted two young engineers to design and build to his idea. One of them was, despite having no formal programming education the best programmer I ever met (*8), who single handily designed the basics for what became one of the largest real time real time applications within the company.
*1 - Even that was a ready a marvel, but that's a different issue.
*2 - It might be important to know that the guys doing this were hardware people and at the time it was strictly forbidden for them to do any programming or even get courses :)) So their base was,'t much how to use a programming environment, but maintenance tools - like patching from the console, so the first program editor was build from patches (the real, hex kind, not source changes) to the OS to add a set of commands that worked like an ED style line editor, which then was used to write the application programs.
*3 - Still today I'm amazed how privileged I was to have learned from that man later on.
*4 - Which is the maximum number of digits a phone number ca have to be routed worldwide, including all escape codes (and yes, I know, this might be relaxed by now).
*5 - Thus all input had to be numeric. Input was terminated by Star (*, caneled by Hash (#)
*6 - 6 bit is more than enough to display all upper case characters, numbers and some symbols, so why spend more?
*7 - think a two desk cubicles sharing a telephone ... yes, that was a common thing in the not so good olde times.
*8 - Again me being the luckiest guy in the universe to have had the chance to work with and learn from for over 20 years. Without, I guess, I would be just an average low payed maintenance engineer firmly believing the C64 is a computer and C a language.