I would not use these devices by choice. I would use a "logic level" MOSFET that has a fantastically stronger current drive and lower rds-on at 5v. Note that it's not just a matter of rds-on, there's a matter of having enough gate voltage to ensure the specified Id is greater than the current used or it will stop being a switch. At 5v, the IRFZ44NIR data sheet I pulled (there are a number of part variations) can drive up to 25A as a switch but one, you want a lot of margin to get a lot rds-on, and two, temp and part variations (particularly in the threshold voltage) can substantially change that chart. Vth is specified to be anywhere 2v-4v. Assuming that chart was for a 3v middle-of-the-road Vth, you get a Vth=4v device and the Vgs-Id relationship shifts down by 1v. Now that part has an Id of only 9amps when cold, though it does go up to about 17 amps when it's at its maximum junction temp. Need more margin!
Take a look at "hexfets" and stuff in a 8-SOP pkg. Like this:
**broken link removed**
Vgs/Id curve shows Id at 5v is insanely more than you will use. Rds =0.003 ohms. You will not need a heatsink, nor can one even be used.
Or
https://www.fairchildsemi.com/ds/FD/FDS7088N7.pdf
Has similar stats. Very strong drive on 5V! Rds-on at 5v, 15 amps is only 1.1mOhms!
The thing to note is these switches have little heat dissipation ability and you can't use a heat sink. With a 1.1 milliohm resistance at Vgs=5v at 15 amps Id, heat is only 1/4 watt. However, switching losses on PWM are going to be VERY important. Losses will basically be proportional to the frequency.
Paralleling the MOSFETs will improve their dissipation and these things are quite cheap and small. The only problem is that the extra capacitance of multiple gates on one PWM pin may slow down the pin's transition due to the capacitive loading. These devices do have a higher gate capacitance at 5v than the IRFZ44.
Many amateurs are initially hesitant to use a SOP-8 pkg since it "feels weak" not to have a heatsink. The only real truth is that the SOP-8 devices will burn out quite fast in some sort of fault condition due to the lower dissipation of the "problem's" heat whereas a big TO-220 with a big heatsink might take it for several seconds or even indefinitely. These SOP-8 devices are plain far superior switches for logic-level job- relatively few TO-220 devices have very strong drive on Vgs=5v.
In a way, TO-220 devices are on the way out as the most common solution, the industry trend is for these far cheaper SMD pkgs, a desire to avoid heatsinks that add many dollars (even hundreds) to the cost, and consumers want smaller devices. Most of the neat new stuff that works so much better is in ultra-small devices.