Adaptive Body Biasing in Strong Body Factor Technologies
With the advent of intelligent sensor nodes in everyday life, low power aspects of system design become more and more important. Adaptive body biasing is a promising methodology to achieve dynamic adaptation of the tradeoff between performance and energy by shifting the threshold voltage of transistors in a digital design. This approach, in combination with a low supply voltage, provides a strong knob to the designer to rapidly shift the circuits operating point from deep sub threshold operation for slow and low leakage retention, to fast, higher performance operation to for short, but demanding tasks. This thesis concentrates on such designs using the deeply depleted channel technology in which body control is particularly effective.
The first part of this thesis is dedicated to strategies and tools supporting the digital design process of circuits using adaptive body biasing. A methodology to compare a standard cell library characterised in different operating points, defined by supply voltage, process corner, temperature, and bias points is presented first.
Next, we present a methodology to exhaustively and rapidly map out the supply-voltage/bias-voltage design space using a heavily pruned cell library. We extract speed and power of a simple example design across the entire design space and show a methodology to scale the characteristics of the small reference up to a more complex design. In a case study this modelling approach achieves an error of less than 1% on the total power relative to an actual characterisation of the full library at the same design point.
The second part of the thesis analyses three chips implementing different biasing schemes in USJC 55nm DDC. The first two were designed by CSEM with components and measurements contributed from this thesis while the third one was entirely designed for this thesis.
The first chip utilises a biasing scheme based on the first order approximation that the circuit speed is proportional to the on-current which can be driven by PMOS and NMOS transistors. This is implemented using an analog control loop, setting both PMOS and NMOS on-currents equal to a reference current provided by current DAC. The SoC characterisation is presented with the objective of identifying suitable operating modes and bias points, including a reliability and retention analysis of the SRAM. A series of ring oscillators constructed from the most common standard cells has also been integrated and provides measurement support for the first part of the thesis.
The second chip extends the Calanda biasing scheme with a secondary regulation loop that is based on an FLL, designed in this thesis, in combination with a configurable standard cell based ring oscillator. The user can directly program a target frequency and the biasing system regulates the current DAC accordingly. We show that this approach effectively overcomes the drawbacks of the current based approach resulting in an effective regulation.
Finally the third chip presents a novel biasing scheme that was designed in this thesis and is tailored toward simplicity. It utilises two constant voltages for the PMOS bias to switch between retention and operation. A charge pump controlled by a standard cell compatible distributed on current balance sensor regulates the NMOS bias such that the on-currents of PMOS and NMOS match. We show that this simple approach, in conjunction with a well chosen operating point, can be efficient across corners.
EPFL_TH10483.pdf
n/a
openaccess
Copyright
11.77 MB
Adobe PDF
0aaf7ff6da612d1a0149417e2aa6fa10