ISSN: 1693-6930
1

AN OPTIMIZED SQUARE ROOT ALGORITHM

FOR IMPLEMENTATION IN FPGA HARDWARE

Tole Sutikno
Department of Electrical Engineering, Universitas Ahmad Dahlan (UAD)
Jln. Prof. Soepomo, Janturan, Yogyakarta 55164, Telp. +62-274-379418, Fax. +62-274-564604
e-mail: [email protected]

Abstrak
Makalah ini menyajikan metode kalkulasi digit-demi-digit yang dioptimalkan untuk
menyelesaikan perhitungan akar kuadrat yang rumit dalam perangkat keras, sebagai algoritma
sederhana yang diusulkan untuk implementasi pada field programmable gate array (FPGA).
Prinsip utama dari metode ini adalah operasi-operasi penggeseran dua-bit dan pengurang-
multipleks, untuk mendapatkan implementasi yang lebih sederhana dan perhitungan yang lebih
cepat. Algoritma ini telah digunakan untuk implementasi akar kuadrat biner tak bertanda 32-bit
dan 64-bit berbasis FPGA secara sukses. Hasil penelitian menunjukkan bahwa metode yang
diusulkan paling efisien sumber daya perangkas keras, bila dibandingkan metode lainnya.
Selain itu, strategi ini dapat dengan mudah dikembangkan untuk implementasi akar kuadrat
yang lebih besar.

Kata kunci: kalkulasi digit-demi-digit, FPGA, Square Root

Abstract
This paper presents an optimized digit recurrence method to solve complicated square
root calculation in hardware, as a proposed simple algorithm for implementation in field
programmable gate array (FPGA). The main principle of proposed method is two-bit shifting and
subtracting-multiplexing operations, in order to achieve a simpler implementation and faster
calculation. The proposed algorithm has conducted to implement FPGA based unsigned 32-bit
and 64-bit binary square root successfully. The results have shown that proposed method is
most efficient of hardware resource compare to other methods. In addition, the strategy can be
expanded to larger number easily.

Keywords: digit recurrence calculation, FPGA, Square Root

1. INTRODUCTION
The direct torque control method (DTC) for AC motors has been well-known that it has
simple structure, especially when compared with field-oriented control (FOC) and good
behaviors such as fast torque response, no requirements for PWM pulse generation, no
requirements for coordinate transformation, no position encoder and current regulators [1-7].
Because of the DTC algorithm is not easy to implement DTC in FPGA hardware, then
the algorithm is usually implemented by serial calculations based on a Microcontroller or Digital
Signal Processing (DSP) [8-11]. These are truly software-based platform and not adequate to
implement a control methods which require very high speed response. As suitable solution, it is
proposed FPGA to support execution very fast tasks [12-14]. One of problem has been
addressed mainly in complicated square root calculation. It is hard to implement on FPGA [15-
17].
The Rough estimation [18], Babylonian method [19], exponential identity [20], Taylor-
Series Expansion Algorithm [21], Newton-Raphson method [22-24], and sequential algorithm (it
is known as digit-by-digit calculation or digit recurrence method) [25-29] have proposed to solve
square root. Nevertheless, the methods above usually do not focus to solve square root
problem in DTC implementation based on FPGA. This paper proposes digit-by-digit calculation
method as a simple strategy to solve complicated square root. The proposed implementation
strategy is different compared to strategies in [25-29]. An optimization is also done by eliminates

An Optimized Square Root Algorithm for Implementation in FPGA Hardware (Tole Sutikno)
2
ISSN: 1693-6930

circuitry that is not needed. It is addressed to support DTC implementation in FPGA hardware,
and in hopes that it gives rise simpler implementation and faster calculation.

2. DIGIT RECURRENCE METHOD

In digit recurrence method, the each digit of the square root is found in a sequence
where it only one digit of the square root is generated at each iteration [29]. It has several
advantages, such as: every digit of the root found is known to be correct and it will not have to
be changed later; if the square root has to expand, it will terminate after the last digit is found;
and the algorithm works for any number base (of course the process depends on number base).

(a) (b)

Figure 1. The example of digit-by-digit calculation to solve square root: (a) restoring algorithm;
(b) non restoring algorithm

Figure 2. The example of using modified non restoring digit-by-digit calculation algorithm to
solve square root

TELKOMNIKA Vol. 8, No. 1, April 2010 : 1 - 8

TELKOMNIKA ISSN: 1693-6930 ■ 3

Two basic techniques have been used to perform the square root operation in
hardware, i.e. restoring and non restoring digit-by-digit algorithm [29]. In restoring algorithm, the
procedure is composed by taking the square root obtained so far, appending 01 to it and
subtracting it, properly shifted, from the current remainder. The 0 in 01 corresponds to
multiplying by 2; the 1 is a new guess bit. The new root bit developed is truly 1, if the resulting
remainder is positive, and vice versa is 0, which the remainder must be restored by adding the
quantity just subtracted. It is different, in non restoring algorithm does not restore the subtraction
if the result was negative. Instead, it appends a 11 to the root developed so far and on the next
iteration it performs an addition. If the addition causes an overflow, then on the next iteration
you go back to the subtraction mode [30].
The example gives to take the binary square root of 01011101 (equivalent with 93
decimal) is shown in Figure 1. A slightly different of non restoring digit-by-digit algorithm in
Figure 1 (b), a modification as shown on Figure 2 can be conducted to give simpler
implementation and faster calculation. In this modification, it only uses subtract operation and
append 01, while add operation and append 11 is not used. This paper adopts this modification
to implement unsigned 64-bit binary square root based on FPGA.

3. PROPOSED SQUARE ROOT ALGORITHM

By eliminate redundant blocks, Samavi [29] has improved classical non-restoring digit
recurrence square root circuit. The Samavi’s circuit is referred to reduce area of non restoring
circuit. However, it still based on constant digit of 01 or 11 and add-subtract as the main building
block (still refer to Figure 1 b). This paper offers a simple alternative solution that it only uses
subtracts operation and appends 01. As consequent, the subtract-multiplex is used as the main
building block (refer to Figure 2). The principle of proposed algorithm can be described as
shown in Figure 3.

Step 0. Start
Step 1. Initialization radicand (the n-bit number will be squared root), quotient
(the result of squared root), and remainder. To calculate square root of a
2n bit number, it needs n stage pipelines to implement the proposed
algorithm.
Step 2. Beginning at the binary point, divide the radicand into groups of two
digits in both direction.
Step 3. Beginning on the left (most significant bit), select the first group of
one or two digit (If n is odd then the first groups is one digit, and vice
versa)
Step 4. Choose 1 squared, and then subtract.
Fist developed root is “1” if the result of subtract is positive, and vice
versa is “0”
Step 5. Shift two bits, subtract guess squared with append 01.
Nth-bit squared is “1” if the result of subtract is positive, and Because
of subtract operation is done
else
Nth-bit squared is “0”, and not subtract
Step 6. Go to step 5 until end group of two digits
Step 7. End

Figure 3. The principle of proposed algorithm to solve square root

Figure 4 shows a simple hardware implementation of the non-restoring digit-by-digit

algorithm for unsigned 6-bit square root by an array structure. The radicand is P
(P5,P4,P3,P2,P1,P0), U (U2,U1,U0) as quotient and R (R4,R3,R2,R1,R0) as remainder. It can
be shown that the implementation needs 3 stage pipelines. The main building blocks of the
array are blocks called as controlled subtract-multiplex (CSM). Figure 5 present the details of a
CSM. Input of the building block is x,y,b and u, and as output is bo (borrow) and d (result). If
u=0, then d<=x-y-b else d<=x.
In Figure 6, the generalization of simple implementation of the non-restoring digit-by-
digit algorithm for unsigned n-bit square root by an array structure is shown. Each row (stage) of

An Optimized Square Root Algorithm for Implementation in FPGA Hardware (Tole Sutikno)
4
ISSN: 1693-6930

the circuit in Figure 6 executes one-iteration of the non-restoring digit-by-digit square root
algorithm, where it only uses subtracts operation and appends 01.

Figure 5. Internal structure of

a CSM block
Figure 4. A simple hardware implementation of the non-restoring
digit-by-digit algorithm for unsigned 6-bit square root

Figure 6. A simple hardware implementation of the non-restoring digit-by-digit algorithm for

unsigned n-bit square root

To optimize hardware resource saving of the implementation above, specialized entities

can be created as building block components. It will eliminate circuitry that is not needed. As
example, the implementation in Figure 6 for unsigned 6-bit square root can be optimized
become as shown in Figure 7 (in this case, the remainder is ignored, because in the DTC drive,
it is not required). The specialized entities A, B, C, D and E are minimized CSM when input
ybu=100, yu=00, u=0, yu=10, and y=0 respectively, and the remainder is ignored. The

TELKOMNIKA Vol. 8, No. 1, April 2010 : 1 - 8

TELKOMNIKA ISSN: 1693-6930 ■ 5

generalization of optimized simple implementation of the non-restoring digit-by-digit algorithm

for unsigned n-bit square root is shown in Figure 8.

Figure 7. Optimized simple hardware implementation of the non-restoring digit-by-digit algorithm

for unsigned 6-bit square root

Figure 8. Optimized simple hardware implementation of the non-restoring digit-by-digit algorithm

for unsigned n-bit square root

4. RESULTS AND ANALYSIS

In the previous sections, optimized simple hardware implementation method of the non-
restoring digit-by-digit algorithm for square root and the difficult task in DTC to calculate square
root were explained. In this section, simulation results of 32-bit and 64-bit square root based on
Altera APEX 20KE FPGA by using method above are presented, as shown in Figure 9. In this
simulation, P is radicand and U is quotient. The results showed that the implementation has
succeeded and worked properly.
The implement 32-bit and 64-bit square root using optimized simple hardware
implementation method of the non-restoring digit-by-digit algorithm with refer to compilation

An Optimized Square Root Algorithm for Implementation in FPGA Hardware (Tole Sutikno)
6
ISSN: 1693-6930

report are needed 256 and 1023 logic element (LE) respectively. The comparison of results
obtained from different implementation method is shown in Table 1.

(a) (c)

(b) (d)

Figure 9. Simulation result of n-bit square root using optimized simple hardware implementation
method of the non-restoring digit-by-digit algorithm: (a) 32-bit in decimal display, (b) 32-bit in
binary display, (c) 64-bit in decimal display, (d) 64-bit in binary display

Table 1. The comparison of logic element usage

LE Usage
No Method
32-bit square root 64-bit square root
1 Classical-NR 1008 4092
2 Reduced-Area-NR 632 2464
3 Modular-NR 624 2468
4 Simple-X-Module 648 2488
5 Proposed 256 1023
Note: Altera APEX 20KE & Xilinx Virtex-E, 1 LC = 1 LE, and 1 CLB = 4 LE [31]

This comparison of LE or logic cell (LC) usage is listed based on references [29] and
[30]. The number of employed LE indicates the size of the implemented circuit “hardware
resource”. Table 1 showed that proposed method is most efficient of hardware resource. Based
on Figure 8, the strategy is very easy to be expanded for larger number to solve complicated
square root problem in FPGA implementation.

TELKOMNIKA Vol. 8, No. 1, April 2010 : 1 - 8

TELKOMNIKA ISSN: 1693-6930 ■ 7

5. CONCLUSION
This contribution presented new digit recurrence method as a proposed simple strategy
for implementation in field programmable gate array (FPGA) hardware mainly to solve
complicated square root. The main principle of proposed method is two-bit shifting and
subtracting-multiplexing operations. The proposed strategy has conducted to implement FPGA
based unsigned 32 bit and 64-bit binary square root successfully. The results have shown that
proposed method is most efficient of hardware resource compare to other methods. The method
also can be expanded to larger number easily, to solve complicated square root problem in
FPGA implementation.

REFERENCES
[1]. Takahashi I, Noguchi T. A New Quick-Response and High-Efficiency Control Strategy of
an Induction Motor. IEEE Transactions on Industry Applications. 1986; IA-22(5): 820-827.
[2]. Depenbrock M. Direct Self Control (DSC) of Inverter-fed Induction Machine. IEEE Trans.
on Power Electronics. 1988; 3(4): 420-429.
[3]. Habetler TG, Profumo F, Pastorelli M, Tolbert LM. Direct Torque Control of Induction
Machines Using Space Vector Modulation. IEEE Transactions on Industry Applications.
1992; 28(5): 1045-1053.
[4]. Zhong L, Rahman MF, Hu WY, Lim KW. Analysis of Direct Torque Control in Permanent
Magnet Synchronous Motor Drives. IEEE Transactions on Power Electronics. 1997;
12(3): 528-536.
[5]. French C, Acarnley P. Direct Torque Control of Permanent Magnet Drives. IEEE
Transactions on Industry Applications. 1996; 32(5): 1080-1088.
[6]. Yong L, Zhu ZQ, Howe D. Direct Torque Control of Brushless DC Drives with Reduced
Torque Ripple. IEEE Transactions on Industry Applications. 2005; 41(2): 599-608.
[7]. Yong L, Zhu ZQ, Howe D. Commutation-Torque-Ripple Minimization in Direct-Torque-
Controlled PM Brushless DC Drives. IEEE Transactions on Industry Applications. 2007;
43(5): 1012-1021.
[8]. Bose BK, Szczesny PM. A Microcomputer-based Control and Simulation of An Advanced
IPM Synchronous Machine Drive System for Electric Vehicle Propulsion. IEEE
Transactions on Industrial Electronics. 1988; 35(4): 547-559.
[9]. Lianbing L, Hexu S, Xiaojun W, Yongqing T. A High-Performance Direct Torque Control
Based on DSP in Permanent Magnet Synchronous Motor Drive. Proceedings of the 4th
World Congress on Intelligent Control and Automation. 2002; 2: 1622-1625.
[10]. Weijie L. Implementation of Direct Torque Control for Permanent Magnet Synchronous
Motor with Space Vector Modulation Based on DSP. 8th International Conference on
Signal Processing. 2006; 4: 101-104.
[11]. Cruz SMA, Toliyat HA, Cardoso AJM. DSP Implementation of The Multiple Reference
Frames Theory for The Diagnosis of Stator Faults in A DTC Induction Motor Drive. IEEE
Transactions on Energy Conversion. 2005; 20(2): 329-335.
[12]. Monmasson E, Cirstea MN. FPGA Design Methodology for Industrial Control Systems: A
Review. IEEE Transactions on Industrial Electronics. 2007; 54(4): 1824-1842.
[13]. Kowalski CT, Lis J, Orlowska-Kowalska T. FPGA Implementation of DTC Control Method
for the Induction Motor Drive. The International Conference on Computer as a Tool
(EUROCON). 2007:1916-1921.
[14]. Colli VD, Di Stefano R, Marignetti F, Scarano M. Design of a System-on-Chip PMSM
Drive Sensorless Control. IEEE International Symposium on Industrial Electronics (ISIE).
2007: 2386-2391.
[15]. Yamin L, Wanming C. Implementation of Single Precision Floating Point Square Root on
FPGAs. IEEE Symposium on FPGA for Custom Computing Machines. Napa, California,
USA. 1997: 226-232.
[16]. Piromsopa K, Aporntewan C, Chongstitvatana P. An FPGA Implementation of A Fixed-
Point Square Root Operation. Int. Symp. on Communications and Information Technology
(ISCIT 2001). ChiangMai, Thailand. 2001: 100-102.
[17]. Lachowicz S, Pfleiderer HJ. Fast Evaluation of the Square Root and Other Nonlinear
Functions in FPGA. 4th IEEE International Symposium on Electronic Design, Test and
Applications (DELTA). 2008: 474-477.

An Optimized Square Root Algorithm for Implementation in FPGA Hardware (Tole Sutikno)
8
ISSN: 1693-6930

[18]. Ercegovac MD. On Digit-by-Digit Methods for Computing Certain Functions. Conference
Record of the Forty-First Asilomar Conference on in Signals, Systems and Computers
(ACSSC). 2007: 338-342.
[19]. Kosheleva O. Babylonian Method of Computing The Square Root: Justifications Based on
Fuzzy Techniques and on Computational Complexity. Annual Meeting of the North
American Fuzzy Information Processing Society (NAFIPS). 2009: 1-6.
[20]. Ligon WB, Monn IG, Stanzione D, Stivers F, Underwood KD. Implementation and
Analysis of Numerical Components for Reconfigurable Computing. IEEE Proceedings
Aerospace Conference.1999; 2:325-335.
[21]. Taek-Jun K, Sondeen J, Draper J. Floating-Point Division and Square Root
Implementation Using A Taylor-Series Expansion Algorithm. 15th IEEE International
Conference on Electronics, Circuits and Systems (ICECS). 2008:702-705.
[22]. Kabuo H, Taniguchi T, Miyoshi A, Yamashita H, Urano M, Edamatsu H, Kuninobu S.
Accurate Rounding Scheme for The Newton-Raphson Method Using Redundant Binary
Representation. IEEE Transactions on Computers. 1994; 43(1): 43-51.
[23]. Allie M, Lyons R. A Root of Less Evil [Digital Signal Processing]. IEEE Signal Processing
Magazine. 2005; 22(2): 93-96.
[24]. Liang-Kai W, Schulte MJ. Decimal Floating-Point Square Root Using Newton-Raphson
Iteration. 16th IEEE International Conference on Application-Specific Systems,
Architecture Processors (ASAP). 2005: 309-315.
[25]. Tchoumatchenko V, Vassileva T, Gurov P. A FPGA Based Square-Root Coprocessor.
Proceedings of the 22nd EUROMICRO Conference Beyond 2000: Hardware and
Software Design Strategies. 1996: 520-525.
[26]. Takagi N, Takagi K. A VLSI Algorithm for Integer Square-Rooting. International
Symposium on Intelligent Signal Processing and Communications (ISPACS). 2006: 626-
629.
[27]. Yamin L, Wanming C. Parallel-Array Implementations of A Non-Restoring Square Root
Algorithm. IEEE International Conference on Computer Design: VLSI in Computers and
Processors (ICCD). 1997: 690-695.
[28]. Xiumin W, Yang Z, Qiang Y, Shihua Y. A New Algorithm for Designing Square Root
Calculators Based on FPGA with Pipeline Technology. Ninth International Conference on
Hybrid Intelligent Systems (HIS). 2009: 99-102.
[29]. Samavi S, Sadrabadi A, Fanian A. Modular Array Structure For Non-Restoring Square
Root Circuit. Journal of Systems Architecture. 2008;54(10): 957-966.
[30]. Comparing Altera APEX 20KE & Xilinx Virtex-E Logic Densities. Altera Corporation. 2010.

TELKOMNIKA Vol. 8, No. 1, April 2010 : 1 - 8