In this article, we introduce two variants of energy efficient M-ary frequency-shift keying (FSK) for low data rate/low power Internet-of-Things (IoT) applications. Both variants, i.e., M-ary direct current (DC)-FSK and M-ary unipolar (U)-FSK are compatible with intensity-modulation and direct detection (IM-DD) implementation of visible light communication (VLC). The two techniques intrinsically differ in the manner of attaining a non-negative signal for intensity-modulation. Mary DC-FSK uses a DC-offset, while, M-ary U-FSK sequentially transmits the positive and the sign flipped negative halves of the bipolar M-ary FSK symbols. The spectral efficiencies of Mary DC-FSK and M-ary U-FSK are augmented by biorthogonal extension of frequency waveforms resulting in 2M-ary biDC-FSK and 2M-ary biU-FSK, respectively. Two optimal maximum likelihood (ML) receiver configurations with different complexities are introduced for M-ary DC-FSK/2M-ary biDC-FSK. Whereas, for M-ary U-FSK/2M-ary biU-FSK, an optimal ML and a suboptimal receiver are proposed. We appraise the performance of these methods in terms of Euclidean distance, bit error rate (BER) in additive white Gaussain noise and time dispersive channels, energy efficiency with respect to spectral efficiency and computational complexity. Simulations confirm that the proposed techniques are more energy efficient than classical M-ary pulseamplitude modulation (PAM) in an absolute sense.