Diodes – Diode-Connected BJT vs Base-Emitter BJT Diode vs Diode

It is there to keep the transistor's current less susceptible to temperature changes.

In the case of Q1:

Suppose that instead of having R1 and D1, Q1base was connected directly to ground.
Emitter current would be: $$ I_{e} = \frac{20V - V_{be}}{R_{2}}$$

You can see Ie is susceptible to variations in Vbe, which has a known dependency on temperature (T), so you might as well express it as: $$ I_{e}(T) = \frac{20V - V_{be}(T)}{R_{2}}$$

But with the diode, if they are matched and thermally bonded: $$ V_{diode}(T) = V_{be}(T) $$

So now: $$ I_{e} = \frac{20V+V_{diode}(T)-V_{be}(T)}{R_{2}} $$ Which simplifies to: $$ I_{e} = \frac{20V}{R_{2}} $$ Independent of Vbe, and its variations with temperature.

The diode is effectively providing the little voltage offset that would be needed to compensate for Vbe changes with T, in order to maintain a constant current.

What you can do is replace a diode with a diode-connected BJT: a two-terminal device made by tying the base and collector together. Sometimes diodes are made this way on integrated circuits. One good reason for actually using a BJT as a diode is that you can closely match another BJT of the same type. This is useful, for instance, if you're trying to build a current mirror with discrete components. If both devices have identical voltage-current curves, and are close together (thermally coupled) then the mirroring is accurate and, free of thermal runaway. Note how current-mirror circuit schematics show a diode-connected BJT, rather than a diode.

By the way, with some resistors, you can turn a transistor into another two-terminal device, the VBE multiplier (a.k.a "rubber diode"). By varying the resistor values you can make the rubber diode exhibit different voltage drops. The diode-connected BJT can be regarded as a special case of the rubber diode, where one resistor is infinite (because there is no connection from base to emitter) and the other is zero ohms (base is tied to collector).

But the question asks, why don't we just use the base-emitter diode in isolation (leaving the collector disconnected). While that will work as a diode, the diode-connected BJT has the advantage that most of the current flows through the collector. In the diode-connected arrangement, the BJT is prevented from saturating, so the current splits between base and collector according to the transistor's beta. If you use only the base and emitter terminals, then you're only pushing base current. Transistors are usually constructed with the assumption that the base current (and hence handling ability) is small compared to collector current.

(The diode-connected transistor doesn't saturate because the collector and base diode cannot forward bias. Transistor saturation occurs when both diodes are forward biased.)