One word: transients.
Inrush current, and surge voltage, to be specific.
Traditional fuses burn in the range of 100us (for ultrafast "semiconductor" and "current limiting" types, at rated peak current), up to 10s or 100s of ms (for a fault current about ten times continuous rated current), or 1 to 10s of sec for slow-blow types, breakers, polyfuses and so on. (Polyfuses span quite a range of size, so that an 0805 size part will open in less than 100ms, but a large leaded part may take tens of seconds.)
It is a desirable
property that a fuse can withstand large surge currents without opening. There are several reasons:
1. Nuisance trips. You don't want a surge (from mains fluctuations, distant lightning, etc.) blowing the fuse every time.
2. Inrush. Many loads gulp huge currents, for durations of a half cycle (e.g., transformers, most SMPSs) to hundreds of cycles (motors). Indeed, a type of resettable time-delay fuse is called a motor starter
. (They're resettable because the current flows through a resistive strip, heating a small solder pot; a gear is secured in the pot. The breaker lever is sprung against the gear teeth. When the solder melts, the gear is able to move, releasing the lever. The mechanism is wide open to see, because the elements often need replacing...)
3. Safety. Nothing else can endure the absolutely tremendous energy delivered by hot and angry pixies charged to 480VAC or more. Even 240V is on the angry side (and, I would suppose, this is reflected by the greater safety measures taken in European mains plugs, compared to the rather lax US style 120V plugs). The peak current ratings of fuses range from 1kA (for LV and residential circuits) to 200kA (for industrial circuits, and maybe more for medium voltage* parts).
*Medium voltage is over 600VAC. 480V is still "low voltage". The practical side is, low voltage doesn't jump out and kill you, you have to touch it first. Medium voltage is on the margin, and includes voltages that jump.
So. Can it be done with semiconductors? Unequivocally, yes! Can it be done as cheaply? Fuck no.
I've designed and built fuse circuits before. At very low voltage, this is trivial: you can buy commercial devices for 5V, 500mA application (USB 2.0) that are tiny surface mount parts. They offer active current limiting and thermal resetting, with a thermal "trip" (on time) in the 2-10ms range, and off-time (recovery) in the 0.1-1s range.
Protected switches are available up to automotive size, e.g., 60V and 10A (suitable for switching 24V loads at as many amps. The current limit is usually inaccurate, landing in the 20-60A range. During a current limit, the thermal trip time is in the 100us range, and recovery in the ~1s range.
Making a power limit much higher than this (i.e., 24V * 60A is almost 1.5kW peak), for any length of time (i.e., 10s of ms, sufficient to start capacitors and motors), quickly becomes impractical. You need sheer volume of silicon now, and the price and size spirals out of control at the same time.
You can employ alternative topologies: instead of limiting current in the transistor, design the circuit to dump faulting power into a dumb load resistor, or something like that. Now you need to switch at fault current levels, but this is still easier than dissipating it.
Suppose you wanted to replace a 15A, 240VAC breaker.
To simplify a little, suppose you make a unidirectional circuit: i.e., it works on DC. This is placed across the DC terminals of a FWB, and the FWB's AC terminals are used to replace the breaker contacts. (Thus, when the circuit is "on", a small voltage is dropped across the FWB, in either direction. When "off", full line voltage is dropped and no current flows.)
Suppose you will design it for 4x current limiting, i.e., 60A. This should be good enough to start most capacitive and motor loads, without also having to handle full short-circuit current (which is around 2kA for a residential circuit, but a breaker employed for this service must safely interrupt 10kA). The nominal peak voltage is about 340V, but we should rate it for at least 600V, to account for lesser transients, which should be clamped with a MOV, which will drop about that much voltage in the process. This is 600V * 60A = 36kVA of switching area, still no small task, but a couple beefy transistors, wired in parallel, will handle that okay. We need a series inductor and a catch diode, to make a boost converter, and a dummy load to burn the dropped power, perhaps a resistor. (If this "fuse" can have a neutral wire routed to it as well, the excess power can be returned to the source, basically making a current limited AC buck converter.)
But perhaps you've already noticed some problems. With clamping at 600V, this device can't be fully off: it will let transients through. Spikes of 2.5kV are largely let through to the load, even when off. That's not very safe!
We could stack a bunch in series, but now the on-resistance goes way up. Or we could use big enough transistors, but 3.6kV IGBTs aren't exactly cheap, they switch slowly (low microseconds), and still have a relatively high voltage drop (about 3V, though that's really not much at all compared to the 3.6kV rating!). (On the upside, the power dissipation rating will be high, too, so we don't have to worry as much about switching efficiency.)
So, it's hard.