An alternate 8T storage-cell that operates down to sub-Vt is also shown in Figure 9b (71). Here, reads are performed on a separate port via the read bit-line (RDBL), so the storage nodes can be isolated, eliminating the read SNM limitation. To maximize density, many cells can share RDBL, and when unaccessed, their leakage currents are gated by properly controlling Buffer–Foot from the array periphery. Last, to enforce the relative strength of M5/6 over the storage devices during a write operation, the cell supply voltage VVDD is appropriately reduced using a peripheral supply driver. Hence, the hazards of parallel off devices imposing bit-line leakage and variation skewing the relative strengths required for writeability are both avoided to achieve full operation to below 0.35 V. The resulting savings in leakage power are over a factor of 20 compared with operation at 1 V.
3.3. Analog-to-Digital Conversion
Because physical biomedical signals are analog, an ADC is required before they can be processed digitally to take advantage of the sophisticated capabilities of a digital signal processor (DSP). Precisely how much processing is done before the ADC is a matter of system-to-system optimization. ADC requirements depend on system characteristics, namely bandwidth and dynamic range, so system optimization must consider ADC power, which can be a significant portion of the total power. As one might expect, the energy per conversion, which is an important metric for ADCs, increases as the dynamic range and sampling rate requirements increase. An empirical figure of merit (FOM) for ADCs normalizes their power consumption to the Nyquist sampling rate, FS, and the dynamic range, expressed as 2ENOB (where ENOB is the effective number of bits output) (72):
An alternate 8T storage-cell that operates down to sub-Vt is also shown in Figure 9b (71). Here, reads are performed on a separate port via the read bit-line (RDBL), so the storage nodes can be isolated, eliminating the read SNM limitation. To maximize density, many cells can share RDBL, and when unaccessed, their leakage currents are gated by properly controlling Buffer–Foot from the array periphery. Last, to enforce the relative strength of M5/6 over the storage devices during a write operation, the cell supply voltage VVDD is appropriately reduced using a peripheral supply driver. Hence, the hazards of parallel off devices imposing bit-line leakage and variation skewing the relative strengths required for writeability are both avoided to achieve full operation to below 0.35 V. The resulting savings in leakage power are over a factor of 20 compared with operation at 1 V.3.3. Analog-to-Digital Conversion Because physical biomedical signals are analog, an ADC is required before they can be processed digitally to take advantage of the sophisticated capabilities of a digital signal processor (DSP). Precisely how much processing is done before the ADC is a matter of system-to-system optimization. ADC requirements depend on system characteristics, namely bandwidth and dynamic range, so system optimization must consider ADC power, which can be a significant portion of the total power. As one might expect, the energy per conversion, which is an important metric for ADCs, increases as the dynamic range and sampling rate requirements increase. An empirical figure of merit (FOM) for ADCs normalizes their power consumption to the Nyquist sampling rate, FS, and the dynamic range, expressed as 2ENOB (where ENOB is the effective number of bits output) (72):
การแปล กรุณารอสักครู่..
