To first order, antenna functionality is a funciton of frequency. @ 100 GHz, the wavelength is 3 mm. A half wave dipole would be 1.5 mm long, not too unreasonable to build in metal. Now, feeding it and generating power at 100 GHz is another issue, and the problem of getting power from the amplifier to the antenna will impact things a whole bunch.
if you want an antenna structure with extremely low losses
and compact size, i would recommend to have a look in the
dielectric resonator antennas.
these antennas can be more expensive than traditional metal
type antennas, but they offer some unique attractive advantages
that otherwise not available with metallic radiators.
THey are especially suited for operating frequencies up to and 100GHz.
See book by Luk ' dielectric resonator antennas' .
At these frequencies printed on-chip antennas will be the tdeal. Bondwire antennas can also be used, or actually a bondwire is capable of radiating at that frequency.
My problem is, bondwire or onchip antenna, they will all be inside the lossy casing of the device. As a result, antenna efficiency will be very low.
I recently left a company where I was working on an antenna that had multiple feeds with bands to nearly 200 GHz. The antennas were offset fed reflectors with corrugated horns illuminating the dish. The waveguides were really small.
As you might expect they were really expensive and not exactly easy to assemble and test. I learned a great deal about CMM's (coordinate measurement machines) while I was there.
A reflector at around 100 GHz is not beyond reality. Tools like TICRA and HFSS would be helpful as well as a good machine shop and precision assembly facility. Testing the thing is also an issue.
One thing for sure is that it would not be inexpensive.