Multidimensional Logarithmic Number System:Hardware Complexity

By -
Hardware Complexity

To provide complexity results for our MDLNS inner product CU, we expand on the inner product processor architecture initially developed for the one-digit 2DLNS [14]. The processor can be used in a systolic array for 1D convolution.

Single-Digit Computational Unit

Figure 84.1 shows the structure of the proposed single-digit CU. Since we do not need to retain the MDLNS representation of the accumulated output, and since the CU is used only in feed-forward

Multidimensional Logarithmic Number System-0166

architectures, we can use the MDLNS domain for the coefficient multiplication and a binary representation for the accumulated output.

The multiplication is performed by small, parallel adders for each of the data and coefficient base exponents. The addition output for each of the b - 1 odd bases is concatenated into an address for a lookup table (ROM). This table produces an equivalent floating-point (FP) value for the product of the odd bases raised to the exponent sum, as shown below:

Multidimensional Logarithmic Number System-0167

We note that since the size of the exponents of each odd base in an MDLNS representation (where there are at least two digits and two bases) can be very small (<4 bits), the maximum address input to the ROM is given by 4 × (b - 1) bits. This is an 8-bit address table for a 3DLNS. For large-dimensional LNS, we can also consider the use of unity approximants to reduce the output of each odd-base adder to the number of bits of the input exponents (or even less if we are willing to accept the increased mapping error). This reduction process stores a small number of unity approximants that can be added in parallel to the output of the odd-base adders. The reduced input to the ROM is selected from these parallel results. The ROM input address size is now reduced by (b - 1) bits.

n-digit Computational Unit

The n-digit computational unit is a simple parallel extension of the one-digit unit. Each of the units computes the binary output for one of the digit combinations. As an example, consider multiplying an accumulating sequence, y, with a coefficient x, z = xy, where

Multidimensional Logarithmic Number System-0168

Clearly there are n2 such units in an n-digit MDLNS.

The parallel outputs are summed at the end of the systolic arrays using an adder tree.

The biggest advantage of the use of more than one digit for the input data and the filter coefficients is that one can obtain extremely accurate representations with very small nonbinary exponents. But the price that has to be paid is that the number of computational channels required is increased to at least 4.

Comments

Post a Comment

Popular posts from this blog

Square wave oscillators and Op-amp square wave oscillator.

By -
Image
Square wave oscillators If Fourier analysis is performed on a square wave it is found that the waveform is composed of many harmonics of the fundamental frequency. This rich harmonic content makes the square wave particularly useful as a quick test of amplifier performance. Examples of possible results are shown. The simplest way to produce a square wave is to make a sine wave oscillator in one of the ways described earlier, then feed the output to some form of squaring circuit such as a Schmitt trigger. This is the method adopted in most commercial sine/square wave generators. There are, however, several circuits for square wave oscillators and a description of the most common follows. Multivibrator Technically the multivibrator is a relaxation oscillator, working on the charging of a capacitor through a resistor. If a negative edge is applied to the circuit in (a), transistor TR 1 turns off the for time taken for the base voltage to return to 0 V. If the voltage step is
Read more

Adders:Carry Look-Ahead Adder.

By -
Image
Carry Look-Ahead Adder As previously stated, the carry, c i +1, produced at the i -th position is calculated as c i +1 = x i · y i ∨ c i ·( x i ⊕ y i ). This means that a carry is generated if both x i and y i are 1, or an incoming carry is propagated if one of x i and y i is 1 and the other is 0. Therefore, letting g i denote x i y i , we have c i +1 = g i ∨ c i p i , where p i = x i ⊕ y i . Here, g i is the formula for the carry generation condition at the i -th position, i.e., when g i is 1, a carry is generated at this position. Substituting g i –1 ∨ p i –1 c i –1 for c i , we get A carry look-ahead adder can be realized according to this expression, as illustrated in Figure 41.10 for the case of four bits [21]. According to this expression, c i +1’s are calculated at all positions in parallel. It is hard to realize an n -bit carry look-ahead adder precisely according to this expression, unless n is small, because maximum fan-in and fan-out restriction is violated at higher
Read more

Timing Description Languages:SDF

By -
Image
SDF Standard Delay Format (SDF) is an ASCII format used to convey timing information between EDA tools. SDF was first developed by Cadence in 1990. Open Verilog International (OVI) approved SDF version 2 in 1993 and SDF version 2.1 and 3 in the following years. Improving SDF version 3 features, resulted in IEEE Standard 1497 [1]. Role of SDF in Design Process SDF can be used in different levels of a design process to annotate timing specifications, including timing data and constraints. Timing constraints are used during the forward-annotation process while timing data are applied during the back-annotation process. Forward-annotation process deals with porting timing constraints to synthesis, floorplanner, layout, and routing tools to meet the required constraints. During the design process, timing data can be back-annotated to analysis tools (including simulators, static timing analysis tools, etc.) to provide more accurate timing representations. SDF Structure An SDF f
Read more