Since x*y gives a nilpotent endomorphism, trace(xy) = 0. If S contains a nilpotent element x, φ cannot be injective. In other words, a finite atol extension of an integral domain has no nilpotent elements.
Let F be such a ring. The maximal ideal of F is nilpotent. Yet there are no nilpotent elements, hence F is a field, and F/K is a finite field extension.
Suppose F contains a purely inseparable element x. When computing the trace of xy, one multiplies by the degree of inseparability, which is the characteristic of K. Thus the trace is 0. With φ(x) = 0, F/K is not finite atol. This is a contradiction, hence F/K is separable.
Let b1 through bn be a basis for F/K. Since F/K is separable, galois theory provides n automorphisms on its normal closure. Let c1 through cn be the embeddings of F into the algebraic closure of K. Let M be the matrix of embeddings of basis elements. In other words, Mi,j = ci(bj. Since the embeddings are linearly independent, M is a nonsingular matrix.
Let W = MT times M, in that order. Verify that Wi,j is the trace of bibj.
Since det(W) = det(M)2 in an integral domain, W is a nonsingular matrix. In particular, W is a nonzero matrix. Find i and j such that bibj has a nonzero trace. Given a nonzero x, let y = bibj/x. Now the trace of xy is nonzero, φ(x) is nonzero, and φ is injective.
Let v be a vector in K and imagine a function that maps b1 through bn onto v1 through vn. Since W is invertible, let u be a vector in K such that u*W = v. Let x be the sum over uibi, and let y be the sum over cibi, using arbitrary coefficients ci. Since trace is K linear, expand the trace of xy. The result is u*W*c, which is the same as v.c, which is our original function applied to y. Therefore φ is onto, and F/K is finite atol.
In summary, S/K is finite atol iff S is the direct product of finitely many finite separable extensions of K.