Although scaling VDD is highly effective for reducing EACT, the circuit delay increases rapidly at very low voltages (as shown in Figure 4) and ELKG increases correspondingly. The opposing energy trends are shown in Figure 7 for a 32b adder, and they give rise to a minimum total energy voltage (66). Importantly, this minimum energy voltage occurs in the sub-Vt regime (i.e., VDD of 0.3–0.4 V) for most practical digital circuits, and the energy savings exceed an order of magnitude compared with the nominal supply voltage. For instance, a scalable fast Fourier transform (FFT) processor, which can provide spectral energy computation for silicon cochleae, operates at a minimum energy voltage of 0.35 V consuming 155 nJ/FFT, over a 15x reduction compared with the nominal VDD energy (67). Using parallelism and pipelining, the corresponding clock frequency of 10 kHz can support the 1 kFFT s−1 throughput requirement of implantable speech processors.