The paper doesn't give the function for the phase noise curve, but by eyeing the graph, this is what I calculated:

Assuming L(f) = k + a/f in the region of integration, (where a,k are constants, and f is frequency)

integral( k+a/f, from f1 to f2 ) = k*(f2 - f1) + a* ln( f2/f1 ), where k=10^-150/10, and a = 10^-140/10 * 12e3, f1=12e3, f2=20e6.

plugging in and convert to dBc:  
10 * log( 1e-15*(20e6-12e3) + 1e-14*12e3*ln(20e6/12e3) ) ~ -76.8 dBc.

My answer is not the same as in the paper, but I think it's pretty close, given that I had to eyeball the function.
But this should demonstrate how to go about integrating the area under a phase noise curve, as a step to calculating the jitter.

Notice that L(f) has units dBc/Hz.  Once you integrate over frequency, you get [dBc/Hz] * [Hz] = dBc.

I hope this helps you.

Regards,
Robert

Buckaroo, your equation for computing jitter from a single phase noise offset seems highly limited to some very specific situation, whatever that may be.  I would advise against making such a blanket statement on using this equation as a rule of thumb for calculating phase noise, especially for more complex phase noise curves such as found in systems with cascaded PLLs.