UC Berkeley

Reducing the energy required per operation is the key to building ultra-low energy systems, and the most effective way of achieving this is to reduce the supply voltage. However, operating CMOS circuits at low supply voltages increases circuit delay, leading to lower circuit performance. In this region, the sub-threshold leakage energy component becomes more pronounced and can even dominate the total circuit energy. Increasing threshold voltages reduces the amount of leakage, but this forces operation in the sub-threshold region where performance and variability become exponentially worse.

The use of the sense amplifier-based pass transistor logic (SAPTL) topology is one approach to reducing the energy per operation. It uses an inverted pass transistor logic (PTL) tree, which inherently has no gain, and hence no power supply connections, eliminating the sources of sub-threshold leakage current. Reducing the threshold voltages of the PTL transistors improves performance, without the leakage current increase associated with conventional static CMOS logic. This reduced threshold voltage also allows the PTL transistors to operate in the super-threshold region, even for very low supply voltages, avoiding the increased delay and variability associated with the sub-threshold operating regime.

Gain is introduced by using drivers and sense amplifiers (SAs) that restore the output voltage swing and provide the appropriate output current to drive the fan-out capacitances. These drivers and SAs are the primary source of sub-threshold leakage, which can be amortized by making the PTL networks complex, and by applying various leakage reduction techniques.

SAPTL-based 90nm test circuits using both synchronous and asynchronous timing schemes have been designed, fabricated and tested. These circuits show leakage and energy characteristics better than the equivalent static CMOS circuits. These test chips also demonstrate rudimentary SAPTL-based design flows using commercially available CAD tools.

Simulation and measurement results of basic synchronous SAPTL building blocks show a 40X-50X reduction in standby current and a 6X reduction in energy when compared to an equivalent CMOS logic block, at the expense of a 10X-30X increase in delay. Operating the SAPTL asynchronously reduces the average delay by 89%. However, adding the necessary handshaking circuitry increases the energy by 31%.

These SAPTL building blocks are used to create a parallel 64-byte asynchronous SAPTL-based CRC generator with a minimum energy point that is 25% lower than that of the static CMOS equivalent, with a 6X delay penalty. Also, due to the nature of the PTL tree, forward-biasing the body of the PTL transistors results in a 10% reduction in delay with no energy penalty.

The advantages of the SAPTL over conventional static CMOS is expected to be more significant as technology continues to scale, where subthreshold leakage continue to prevent supply voltages from being aggressively scaled.