Newton's method vs gradient descent

Compared with plain gradient descent, what does Newton's method use, and what's the trade-off?