You should be able to engrave permanent images on metal or glass using a ceramic or metal stylus on a flexing piezoelectric arm with no joints, somewhat similar to the needle of an STM or AFM. Aluminum oxide (sapphire) is probably the easiest ceramic to use, although it provides no electrically-conductive feedback about contact; metal (hardened tool steel, say), or conductive ceramics like tungsten carbide, would remedy that shortcoming, allowing sensitive calibration of engraving depth.
The total movement range of such a setup is likely very small, in the neighborhood of a millimeter, but it can potentially micro-forge the surface of the metal with sub-nanometer resolution, enabling the direct engraving of holograms. I’m not sure what it will do on glass, but I think it’s possible to get it to make scratches rather than just break the glass.
To take a random example, the American Piezo catalog lists a “PSt 150/5x5/7” osi-type piezoelectric stack actuator of 5 mm × 5 mm, 9 mm long, with a maximum stroke of “13/9” μm (not sure what’s up with the two numbers), 800 nF capacitance, resonant frequency of 100 kHz, 120 N/μm of stiffness, 1600 N of blocking force, with a maximum load of 2000 N, operating from -30 V to +150 V. If you hook up three of these things in parallel to a chunk of metal with a grain of aluminum oxide on its tip, they could jam that grain 9 μm into a bit of aluminum, titanium, glass, or even steel, pull it back out, move it into a different position, jam it back in, and repeat, at 25 kHz (two octaves below resonant) without any difficulty.
If the chunk of metal has a 10:1 aspect ratio, for example if the piezo actuators are attached to it 20 mm apart and it’s 200 mm long, then you can wave that little grain of sapphire back and forth by 90 μm, almost the width of a hair. Some kind of flexure-lever arrangement to amplify this by another factor of 10 might be a good idea.
You could presumably engrave data on a little spot at about 50 kbps in this way. (They also offer flat chip actuators with a shorter stroke but much higher resonant frequency, like 500 kHz.) But then you would probably need some kind of repeatable positioning apparatus to engrave over a wider area.
The pressure generated is enormous. Supposing that the tip is up to 25 μm in diameter, 1600 N spread over 490 square microns is 3.3 terapascals. Different metals have yield stresses in the range of 90 MPa (copper) to 2.5 GPa (piano wire), with around 500 MPa being normal; this is 6600 times lower. (Actually annealed aluminum is down around 15–20 MPa.) Sapphire’s ultimate tensile strength is only around 1.9 GPa, and diamond only 2.8.
(Oops, actually the force and pressure from the actuator is potentially 3× that if you have three parallel actuators.)
So you really could find a way to gear these actuators up by a factor of 1000 or so, it would give you a stroke of 9 mm with potentially nanometer precision, and still plenty of force to engrave the surface. (I suppose this is why normal STMs use bending actuators.) If you could still manage 50 kbps, although this seems more dubious, you could engrave a 9 mm square area at 500-nm resolution at 0.0125 mm² per second, filling the whole area in about two hours. This is a good timescale.
Unfortunately, at least the Physikinstrumente devices that gear up piezoelectric actuators in such a fashion have much lower resonant frequencies, like 150 Hz.
As an alternative, maybe you could use electromagnetics, which can also reach up into the MHz range.