Basically it's a graphics terminal whose entire operations are controlled from the server.
Bootp, now DHCP, was originally developed to allow diskless devices to boot an OS from a network server. Thus Sun's ELC diskless workstation, circa 1988, ran a full SunOS implementation locally, but loaded it from a central server. Since disk was very expensive at the time this looked like a great way to save money. Unfortunately it was never really successful because it generally seemed to combine the worst of both worlds: having a general purpose local OS you couldn't modify and save while also having to load and save files over a relatively slow network.
Ten years later Sun tried this again with the JavaStation -a microsparc based machine that loaded and ran the Java virtual machine locally. I bought two, spent a week getting them to work, and sent them back because it managed to combine the worst of three worlds: java, remote boot, and no user control.
Diskless workstations are still common - in fact, that's what most winterms are- but the distinguishing characteristic is no longer the network OS boot since many now boot Linux or Windows CE from ROM or a local disk. Instead, the characteristic that distinguishes the smart display from the diskless workstation is that the latter has the ability to run local programs concurrently with server programs.
Today there are at least a dozen companies making these and they all have exactly the same problem: they're clients in a client-server world. Even the ones that boot Linux are usually deployed as cheaper Windows boxes with a different brand name and administrative structure - that doesn't make them useless, of course, but they're not a general purpose solution either.
The Linux terminal server project, for example, uses this approach - and I just recommended it to an ask bloggie volunteer who needed a way to make some older PC gear work well for his clients.
The first real smart displays, NCD's X-terminals, also booted an OS from the server - but one limited to running the display. As a result the late eighties and early ninties models, things like the 19C and HMX series, were almost 100% reliable - in fact I bought a a batch of 30 used ones with 21 inch screens in 1990 and not one of them failed during ten years of nearly continuous power-on.
The next generation did move toward running some applications on the terminal CPU - this Explora Pro, for example, had the ability to run Mozilla from ROM and marked the beginning of the end for NCD which went on to make winterms and losses.
I mention the Explora in particular here because it was the only NCD series with which, having run hundreds of the things for years, I've ever had any problem -and the story is worth sharing for the laugh value.
See that little black box, about the size of a thin paperback? That's where the processor and memory are. So I had one user whose Explora kept mysteriously failing - he'd be working away when the screen would fade, colors would change, or the picture would distort. He could show it to me after it happened, but no matter how long I sat in his office it would never happen while I was there, and switching the boxes produced the odd effect that the defective one would work perfectly, and the good one, now in his office, would start to fail.
Finally, of course, I caught him - he was putting the processor box on the radiator, dropping a sweater over it, and then putting it back on his desk while calling me. I never did understand why he did it, but this was the only failure incident (other than an office manager who dropped a monitor after discovering it weighed 85 pounds - and a secretary who shipped it across the country by Fed Ex overnight for replacement!) I ever had to deal with.
Sun's Sun Ray carries this evolution one step further. The Sun Ray is not an X-11 device, it literally is just the information display and input hardware for an X-client that runs on the server. That's why you can take a baseball bat to one while someone's using it, and have that person continue typing on another one with no more hassle than signing in again.
One good way to see what the difference amounts to is to think about it in terms of a voting machine for use in elections. Put any kind of local OS on the machine registering the vote - whether that's a PC or Winterm doesn't make any difference - and you cannot ever hope to prove the device secure against tampering. No matter what fancy technologies or processes you invent, any real expert hauled into court to testify about the possibility of vote tampering at source would have say "NO" when asked if tampering is impossible.
In contrast, set this up using Sun Rays, and those experts, facing the same question, would be saying "Yes" since tampering at source is proveably impossible because there's nothing there to tamper with.
So what's a smart display? in the hardware sense it's an information display and input device - nothing more. Sun's ads to the contrary: it's not a client, not a thin client, not even an ultra-thin client. A smart display it collects keystrokes, mouse movements, whatever - and displays the server's response. Nothing else.
In the more general sense, of course, it's part of a deployment philosophy focused on giving the user control, and then getting out of the way That's what NCD's early X-terminals allowed you to do and that's what Sun Rays do today - in both cases the power comes from the classic simplicity of Unix programming: these things do exactly one thing, nothing else, and therefore do it as well as possible.