DESIGN OF EFFICIENT AI ACCELERATOR BUILDING BLOCKS IN QUANTUM DOT CELLULAR AUTOMATA (QCA)

2022-7
Ahmed, Ahmed Mamdouh Mohamed
Quantum-Dot Cellular Automata (QCA)-based digital circuit design have several advantages over CMOS, including higher intrinsic switching speeds up to Terahertz, lower power consumption, a smaller circuit footprint, and higher throughput due to the intrinsic signal propagation scheme’s compatibility with pipelining. As a result, QCA is an excellent choice for applications like Artificial Intelligence (AI)-accelerators, where real-time energy efficient performance is required at a low cost. This Thesis investigates a novel QCA design approach based on an optimal mix of Majority and NAND-NOR-INVERTER (NNI) gates with the USE (Universal, Scalable, Efficient) clocking scheme for latency and energy consumption improvements to fundamental building blocks in AI-accelerators, such as multipliers, adders, accumulators, and SRAMs. The proposed approach was used to design the standard 4×4 Vedic multiplier, which resulted in a 62.8% reduction in cell count, 82.2% reduction in area, and 71.2% reduction in latency over state-of-the-art. The proposed 8-bit PIPO register was simulated to have an 83% reduction in cell count, 94.5% reduction in area, and 94.6% reduction in latency. The proposed SRAM cell architecture is expected to achieve comparable performance improvements as the sub-blocks, such as the D-Latch, which has been modeled to achieve a 44.4% reduction in cell count, a 50% reduction in both area and latency, and a 73% reduction in energy dissipation. The results of this research can be directly applied to low-cost, high-throughput, energy efficient AI-accelerators that could potentially deliver orders of magnitude better energy delay characteristics than their CMOS counterparts, and significant improvements over current QCA implementations.

Suggestions

A Low-power capacitive integrated CMOS readout circuitry for high performance MEMS accelerometers
İncedere, Osman Samet; Akın, Tayfun; Eminoğlu, Selim; Department of Electrical and Electronics Engineering (2013)
This thesis presents a low power capacitive integrated CMOS readout circuitry for high performance MEMS accelerometers. It proposes a linearized model of the complete closed loop accelerometer system, which makes easier of designing and analyzing the system. Designed readout circuitry offers low noise, wide dynamic range and high linearity system with very low power consumption. Designed readout circuit includes proportional integral (PI) controller circuit, which significantly decreases the proof mass defl...
Improving the efficiency of microwave power amplifiers without linearity degradation using load and bias tuning in a new configuration
Ronaghzadeh, Amin; Demir, Şimşek; Department of Electrical and Electronics Engineering (2013)
Advanced digital modulation schemes used in the wireless applications, result in the modulated RF signals with high peak to average power ratio which requires linear amplification. On the other hand, the demand for a longer talk time with less battery volume and weight, especially in hand-held radio units, necessitate more power efficient methods to be utilized in power amplifier design. But improved linearity and efficiency have always been contradicting requirements demanding innovative power amplifier and...
Linearization of RF power amplifiers with memoryless baseband predistortion method
Kolcuoğlu, Turusan; Demir, Şimşek; Department of Electrical and Electronics Engineering (2011)
In modern wireless communication systems, advanced modulation techniques are used to support more users by handling high data rates and to increase the utilization efficiency of the limited RF spectrum. These techniques are sensitive to the nonlinear distortions due to their high peak to average power ratios. Main source of nonlinear distortion in transmitter topologies are power amplifiers that determine the overall efficiency and linearity of the transmitter. To increase linearity without sacrificing effi...
A Low-power memory CMOS integrated circuit for image sensors
Üstündağ, Mithat Cem Boreyda; Akın, Tayfun; Department of Electrical and Electronics Engineering (2015)
This thesis presents a low power SRAM block implemented in a 0.35 μm CMOS technology for imaging applications to be used inside a digital image processor ASIC (Application Specific Integrated Circuit). The SRAM structure is designed to be fast enough to store all the image data fed by a large format readout circuitry such as VGA (640x512), while requiring low power consumption. The low power consumption is a very critical requirement of such circuit, as the circuit will eventually be used in an embedded pla...
Comparison of multi-phase interleaved boost converters with various coupled inductor topologies
Gökmen, Raşit; Bostancı, Emine; Department of Electrical and Electronics Engineering (2022-2)
Multi-phase interleaved boost converters are widely used in high power applications thanks to their high efficiency, high power density, increased output power capability and low input current ripple features. However, increasing the phase number results in an increase in the overall volume of the converter. In order to reduce the volume of the converter, inductors in multi-phase interleaved boost converters can be coupled. For a two-phase interleaved boost converter, loosely-coupled inductor (LCI), close-c...
Citation Formats
A. M. M. Ahmed, “DESIGN OF EFFICIENT AI ACCELERATOR BUILDING BLOCKS IN QUANTUM DOT CELLULAR AUTOMATA (QCA),” M.S. - Master of Science, Middle East Technical University, 2022.