Abstract:
Ternary content addressable memory (TCAM) is a storage device which can store large lookup table with bit-level masking capability. It compares an input search word against the table of stored words and returns the addresses of the matching words. The single clock cycle throughput makes TCAMs attractive for applications such as high-speed packet forwarding and classification in network routers. But, TCAMs consume large amount of energy during table lookup. Reduction of this energy is the most critical challenge faced by the TCAM designers. Energy consumption in TCAMs is dominated by frequent charging and discharging of highly capacitive match lines (MLs) and search lines (SLs). This work proposes four circuit level techniques to reduce TCAM ML energy consumption. 180nm 1.8V CMOS logic has been used for design and simulation of the proposed techniques implemented for internet protocol version 6 (IPv6) packet forwarding. The first scheme uses charge sharing principle. This technique achieves an energy saving of at least 10% compared to popular current race (CR) scheme of similar speed. Comparison with existing low-energy and high-speed techniques employing similar charge sharing principle shows the superiority of the proposed scheme in terms of both search time and energy consumption. In order to achieve further energy saving, a MLSA which uses positive feedback using a single transistor as the feedback element has been proposed. The result is low design complexity and high energy saving in the range of 36% to 45%. But, voltage margin degrades in this case. To get improved voltage margin, this feedback technique has been combined with selective precharge which increases implementation complexity to some extent. The combined technique offers at least 42.5% and at most 55% energy saving. Finally, a dual feedback technique which uses two positive feedback mechanisms has been proposed. This technique provides the highest search speed, voltage margin and energy saving (83.8%) while maintaining low peak power and low implementation complexity. But, multiple feedback mechanisms make the MLSA circuit operation sensitive to process-voltage-temperature variations which can be nullified by adjusting an external bias voltage.
