MAT: Processing In-Memory Acceleration for Long-Sequence Attention | IEEE Conference Publication | IEEE Xplore