An adiabatic charge pump based charge recycling design style

A typical CMOS gate draws charge equal to C[subscript L]Vdd2 from the power supply (Vdd) where C[subscript L] is the load capacitance. Half of the energy is dissipated in the pull-up p-type network, and the other half is dissipated in the pull-down n-type network. Adiabatic CMOS circuit reduces the dissipated energy by providing the charge at a rate significantly lower than the inherent RC delay of the gate. The charge can also be recovered with an RLC oscillator based power supply. However, the two main problems with adiabatic design style are the design of a high frequency RLC oscillator for the power supply, and the need to slow down the rate of charge supply for lower energy. This reduction in speed of operation renders this adiabatic technique inapplicable in certain situations. A new approach incorporating an adiabatic charge pump that moves the slower adiabatic components away from the critical path of the logic is proposed in this work. The adiabatic delays of a charge pump are overlapped with the computing path logic delays. Hence, the proposed charge pump based recycling technique is especially effective for pipelined datapath computations (digital signal processing, DSP, is such a domain) where timing considerations are important. Also the proposed design style does not interfere with the critical path of the system, and hence the delay introduced by this scheme does not reduce the overall computational speed. In this work, we propose one implementation schema that involves tapping the ground-bound charge in a capacitor (virtual ground) and using an adiabatic charge-pump circuit to feed internal virtual power supplies. As the design relies on leakage charge to generate virtual power supplies, it is most effective in large circuits that undergo considerable switching activity resulting in substantial charge tapping by the proposed scheme. The proposed method has been implemented in DSP applications like FIR filter, DCT/IDCT filters and FFT filters. Simulations results in SPICE indicate that the proposed scheme reduces energy consumption in these DSP circuits by as much as 18% with no loss in performance, paving way for a new approach towards conserving energy in complex digital systems