transconductance amplifier
Maybe you can use gm subtraction: two diff pairs with one gm = 100 nS and the other 90 nS connected in opposite polarity. 100nS is easier to implement than 10 nS, for sure... however, this will typically require some work to guarantee gm over PVT..
And on the inversion thing: the stronger inverted the device, the lower its gm/Id --> the more inefficiently the current is used; that's what you want in this case.
If your system lets you, you could also consider duty cycle operation of a larger gm (1 uS 1% of the time, for example)