The document proposes designs for approximate multipliers using approximate compressors to reduce power consumption with small errors. It describes a significance-driven logic compression scheme that uses more accurate compressors for higher significance weights and less accurate compressors for lower significance weights. It also proposes using high-order approximate compressors for middle significance weights to reduce logic and delay. Simulation results show the proposed approximate multiplier has shorter delay and lower power than an exact compressor multiplier while maintaining high accuracy.
The document proposes designs for approximate multipliers using approximate compressors to reduce power consumption with small errors. It describes a significance-driven logic compression scheme that uses more accurate compressors for higher significance weights and less accurate compressors for lower significance weights. It also proposes using high-order approximate compressors for middle significance weights to reduce logic and delay. Simulation results show the proposed approximate multiplier has shorter delay and lower power than an exact compressor multiplier while maintaining high accuracy.
The document proposes designs for approximate multipliers using approximate compressors to reduce power consumption with small errors. It describes a significance-driven logic compression scheme that uses more accurate compressors for higher significance weights and less accurate compressors for lower significance weights. It also proposes using high-order approximate compressors for middle significance weights to reduce logic and delay. Simulation results show the proposed approximate multiplier has shorter delay and lower power than an exact compressor multiplier while maintaining high accuracy.
• Inexact (or approximate) computing is an attractive paradigm for digital
processing at Nano-metric scales. Inexact computing is particularly interesting for computer arithmetic designs. • Our proposed model improvises on 16 x 16 and 8x8 multiplier design. • These designs rely on different features of compression, such that imprecision in computation (as measured by the error rate and the so-called normalized error distance) can meet with respect to circuit-based figures of merit of a design (number of transistors, delay and power consumption). • Two different schemes for utilizing the proposed approximate compressors are proposed and analyzed for a proposed approximate multiplier can achieve both low power and high accuracy. SIGNIFICANCE Significance Driven Logic Compression. • According to the significance, different weights use different compressors (i.e., counters) to accumulate their product terms. • The higher significance weights use accurate 4:2 compressors, the middle significance weights use near-accurate compressors, and the lower significance weights use inaccurate compressors. As a result, the power consumption can be reduced with a small error. High Order Approximate Compression. • For the middle significance weights, we use high-order approximate compressors (e.g., 8:2 compressor) to reduce the logic of carry chains. As a result, both the delay and the power can be greatly saved. • To the best of our knowledge, the proposed design is the first work that utilizes high-order approximate compressors in the approximate multiplier design. EXSITING MODEL • Carry Model For 5:2 Compressor: • The Approximation of Sum Here, we study the approximation of the logic of Sum output. • Conventionally, the tree of XOR gates are used to produce the output Sum. • However, compared with other logic gates, XOR gate often has larger design overheads. We use the logic gates in SAED 32nm cell Figure 1: Representing the carry model for 5:2 compressor library as an example. Table tabulates the comparisons among OR gate, NOR gate, XNOR gate, and XOR gate. • The existing design ensures the design with 5:2 compressor design with sum and carry equations with the logical elements mention in figure 2. For accurate and exact scenario XOR is replaced with XNOR and OR is replaced with NOR.
Figure: Sum of 5:2 compressor. (a) Accurate (b) Our
approximate. COMPRESSORS AND ITS FORMULATION Proposed Higher order compressor 7:3: • The above design model for 7:3 is comprised with a four set of 3:2 compressor and single 4:2 with two 3:2.Each set of structure is observed with the no of inputs are consider as the requirement. • 1. Incase of 4:2 and 3:2 we have only two 3:2 models are connected with one 4:2 model. • 2. Similarly for one 6:3 and one 3:2 model is required for implementing the 7:3 compressor. ….CONTD • The speed of addition is limited due to the time taken by the carry signal to propagate through the adder. • The approx. regular adder was introduced to mitigate the problem of carry propagation delay by independently generating multiple carries and then selecting the correct sum and carry outputs depending on the value of previous carry. • As previously discussed, this type of was not area efficient due to the use of pair of Adder structure to impart the design capabilities to produce the final sum and carry output. ….CONTD A new approximate adder is presented. This adder operates on a set of pre-processed inputs. The input pre-processing (IPP) is based on the interchange-ability of bits with the same weights in different addends. For example, consider two sets of inputs to a 4-bit adder: i)A= 1010,B= 0101and ii)A= 1111,B= 0000. Clearly, the additions of i) and ii) produce the same result. In this process, the two input bits AiBi= 01are equivalent toAiBi= 10(with i being the bit index), because of the interchangeability of the corresponding bits in the two operands. ….CONTD • The basic rule for the IPP is to switch Ai and Bi if Ai= 0andBi= 1(for an yi), while keeping the other combinations (i.e.,AiBi= 00,10and11) unchanged. By doing so, more1’s are expected in Aand more0’s are expected in B. If ̇Ai ̇Bi are the ith bits in the pre-processed inputs, the IPP functions are given by: • A_x=A_i+ B_i.. (1) • A_xi=A_i* B_i.. (2) RESULTS AND DISCUSSION • As per the design criteria we have observed a direct variation of the simulated values from the input changes from the values of • 1. 12 and 6 • 2. 10 and 10 • 3. 10 and 15 • Each set of values are representing one of the approx. and accurate design ensuring the as per the design approach. COMPARISON TABLE CONCLUSION • A novel approximate multiplier design is proposed using a newly designed approximate adder. • On a statistical basis the proposed multiplier has a very small error distance and thus a high accuracy. • Simulations have shown that the proposed design has a shorter critical path delay and a significantly lower power consumption compared to an exact compressor multiplier. • It also uses a configurable error recovery that can produce more accurate results than other state-of-the-art approximate multipliers. ADVANTAGES AND SCOPE • The power saving may be increased if the following criterion is considered in the future low power VLSI design. • Number of bits considered may be increased in the encoding scheme • Power can be reduced by improving the partial product compression ratio.