For highly integrated implantable applications requiring microwatt power levels, switched capacitor voltage converters are an effective approach. Switched capacitor converters can be integrated on the same silicon chip with other processing and communication circuits without external components, allowing for a compact system realization. As the output voltage is dictated by fixed capacitor ratios, the majority of capacitive voltage converters have a limited, discrete set of output voltages. The implementation of dynamic voltage scaling (DVS), to reduce power consumption, requires variable supply voltages. Switched capacitor converters can be enhanced with additional switches and capacitors to be able to realize sufficient granularity of output voltages for DVS (34). Switched capacitor converters are used in implantable pacemakers to generate the 5 V supply that is used to stimulate the heart muscle (1, 35).
At power levels in the tens of milliwatt range and above, external inductors and capacitors are usually required to realize efficient voltage converters. At these power levels, inductor/transformer-based converter architectures are preferred to switched capacitor converters due to their high efficiencies and ability to generate arbitrary output voltages. Inductor-based converter architectures are simple to realize, requiring only a few digital switches, an output filter, and control circuitry (36). Transformer-based switching converters are used in cardioverter-defibrillators, which use a buck-boost flyback converter to generate 750 V pulses to shock the heart (37). Ultimately, with proper circuit design, both switched capacitor converters and inductor/transformer-based converters can efficiently power portable and implantable biomedical devices.
3.2. Ultralow-Power Signal Processing
The types of processing operations required for biomedical devices include filtering, spectral analysis, correlation, threshold/envelope detection, modulation, and data compression. Further, in all cases, the transfer functions must adapt to the perceptions and responses of individual users. As a result, a high degree of programmability in the defining parameters is critical. Generally, two processing domains exist: analog and digital. Hybrid approaches, where the strengths of one processing domain are used to assist the other, also exist for specialized applications (38, 39). Broadly, however, analog signal processing is governed by the rich input-output characteristics of transistors. On one hand, as described in Section 3.2.1, certain complex computations can be performed very efficiently by exploiting these (40); however, sensitivities to environment, biasing, noise, and variation limit their dynamic range. Digital signal processing, on the other hand, quantizes the signal to margin against these sensitivities, but requires additional hardware to process the quantization residue. Here, the need to convert physical analog signals into digital signals with the appropriate resolution, via analog-to-digital conversion, is unavoidable. Because this can be a high-power operation, analog preprocessing should be considered to reduce the ADC dynamic-range or sampling rate requirements.
3.2.1. Low-power analog signal processing.Critical considerations for analog circuits are the power cost of increasing dynamic range and the inconsistency in precise device behavior. The human ear, for instance, can detect sounds ranging from minute air vibrations, on the order of a tenth of an atomic diameter, to noises at the threshold of pain, representing over 110 dB of dynamic range. In analog circuits, however, large-signal excursions alter device operating conditions, resulting in nonlinear distortion, and small-signal excursions are indecipherable due to fundamental device noise. Unfortunately, in advanced technologies, the linear range is decreasing rapidly due to voltage limitations that the fine device features can withstand. Simultaneously, increasing the signal-to-noise ratio (SNR) fundamentally requires a quadratic increase in power. Consequently, selective device biasing and circuit topologies are required to perform analog computations efficiently. Moreover, as many biosignals of interest are at low frequencies, circuits are highly susceptible to 1/f and popcorn device noise. Techniques, such as chopper stabilization and correlated double sampling (41), which are used to cancel device parameter offsets, also help to manage these (42).
Sub-threshold operation.The transconductance behavior of a MOSFET, where an input voltage, VGS, generates an output current, ID, is critical to analog circuits. The threshold voltage, Vt, loosely separates the two regimes of VGS, where ID, for an NMOS, is given by Equations 2 and 3 (43), respectively: