Nearest neighbor (NN) search is an essential operation in many applications,\nsuch as one/few-shot learning and image classification. As such, fast and\nlow-energy hardware support for accurate NN search is highly desirable. Ternary\ncontent-addressable memories (TCAMs) have been proposed to accelerate NN search\nfor few-shot learning tasks by implementing $L_\\infty$ and Hamming distance\nmetrics, but they cannot achieve software-comparable accuracies. This paper\nproposes a novel distance function that can be natively evaluated with\nmulti-bit content-addressable memories (MCAMs) based on ferroelectric FETs\n(FeFETs) to perform a single-step, in-memory NN search. Moreover, this approach\nachieves accuracies comparable to floating-point precision implementations in\nsoftware for NN classification and one/few-shot learning tasks. As an example,\nthe proposed method achieves a 98.34% accuracy for a 5-way, 5-shot\nclassification task for the Omniglot dataset (only 0.8% lower than\nsoftware-based implementations) with a 3-bit MCAM. This represents a 13%\naccuracy improvement over state-of-the-art TCAM-based implementations at\niso-energy and iso-delay. The presented distance function is resilient to the\neffects of FeFET device-to-device variations. Furthermore, this work\nexperimentally demonstrates a 2-bit implementation of FeFET MCAM using AND\narrays from GLOBALFOUNDRIES to further validate proof of concept.\n