Efficient design of a quantum absolute-value circuit using Clifford+T gates

Current quantum computers have a limited number of resources and are heavily affected by internal and external noise. Therefore, small, noise-tolerant circuits are of great interest. With regard to circuit size, it is especially important to reduce the number of required qubits. Concerning to fault-tolerance, circuits entirely built with Clifford+T gates allow the use of error correction codes. However, the T-gate has an excessive cost, so circuits with a high number of T-gates should be avoided. This work focuses on optimising in such terms an operation that is widely used in larger circuits and algorithms: the calculation of the absolute-value of two’s complement encoded integers. The proposed circuit halves the number of required T gates with respect to the best circuit currently available in the literature. Moreover, our circuit requires at least 2 qubits less than the other circuits for such an operation.


Introduction
Quantum computing [1] is a new computational paradigm that explicitly uses quantum phenomena such as superposition, interference and entanglement to achieve speed-ups over what is possible with classical computers in certain tasks. For instance, the famous algorithm for integer factorization introduced by Peter Shor [2] is exponentially faster than the best classical algorithm for the same task that we have at our disposal. Another provable speed-up is the one provided by Grover's algorithm for black-box search problems [3], achieving a quadratic speed-up over any algorithm that only uses classical resources.

3
Efficient design of a quantum absolute-value circuit using… There are several realizations of quantum computing, but the most popular oneand the one currently implemented in most quantum computers -is the quantum circuit model [1]. This model is analogous to the (classical) digital circuit model, but it uses quantum bits (or qubits) instead of bits, quantum gates (or unitary transformations) instead of digital gates and, in addition, it needs explicit measurements to obtain results.
Quantum algorithms, including Shor's factoring method and Grover's search algorithm, usually depend on circuits that implement arithmetical primitives such as modular addition, multiplication and exponentiation, or, for instance, sign and absolute value computations in two's complement [4]. In principle, these operations can be directly adapted from their classical counterparts with linear overhead, but reducing the size and depth of quantum circuits is extremely important to mitigate the effects of noise and imperfect gate realization.
For these reason, a very active line of research in the quantum computing field [5][6][7][8] is devoted to producing quantum circuits for arithmetical primitives that optimize some important metrics [9][10][11] such as number of gates, T-count, number of two-qubit gates, ancillary qubits and garbage outputs.
In this paper, we follow this line of research in the particular case of the absolute value computation for integer numbers in two's complement representation. Computing the absolute value of such numbers is an essential operation in quantum algorithms with applications in fields as diverse as audio signal processing, protein structure prediction, or stock market indicators [12][13][14]. In terms of relevance, the calculation of the absolute value can account for from 10% of these algorithms to more than 30% , as in the case of quantum audio signal inversion, addition, or delay operations [12]. The availability of efficient circuits for the absolute value computation, as well as for the rest of the operations involved in these algorithms, will increase their overall efficiency.
In this work, two fundamental objectives are pursued. The first is to achieve a circuit built exclusively from Clifford+T gates to achieve a fault-tolerant implementation. As a second objective, a cost-effective implementation is sought to reduce costs compared to currently available circuits. Combining these two objectives is not trivial, as will be explained in the next section. However, the obtained results will show that the proposed circuit manages to achieve both objectives.
The rest of the paper is organized as follows. Section 2 introduces the basic concepts about quantum circuits, gates, and metrics necessary for the understanding of this work, as well as the methodology used for its development. Section 3 presents the proposed circuit in detail, indicating all the necessary steps for its reconstruction. Section 4 presents the obtained results, as well as a comparison between the proposed circuit and the most relevant circuits available in the state-of-the-art. Finally, the last section presents the conclusions.

Quantum circuits, gates, and metrics
Quantum circuits are built using qubits and quantum gates that modify the state of the qubits. In classical circuits, there are only two logic gates that allow working 1 3 on a bit: the NOT gate and the identity gate [15]. At the two-digit level, there are not many more gates available. However, there are infinite quantum gates that allow working on one qubit (and therefore also infinite possibilities for cases with two or more qubits) [16]. Fortunately, there are sets of quantum gates that allow to approximate the infinite possible quantum gates. One of these groups, but not the only one, is the so-called Clifford+T group. The Clifford group, generated by the H, S, and CNOT gates, has a wide variety of applications thanks to its feature of being able to convert any Pauli operation into another Pauli operation. Nevertheless, it is not universal on its own, but it needs the T gate to form the Clifford+T group and, become a universal set [1,16].
One of the goals of this work is to get a circuit implemented exclusively using Clifford+T gates. However, being a universal group is not the only reason for its choice. Current quantum computers are very sensitive to internal and external noise. It has been shown that building circuits using only Clifford+T gates will make such circuits fault-tolerant thanks to the use of error correcting codes [17,18]. Nevertheless, there is an associated problem. The cost of the T gate is much higher than that of the Clifford gates, up to 100 times higher. The T-gate is so expensive that a current trend in the quantum circuit literature is to measure the cost of quantum circuits in terms of the T-gate. This metric -the number of T-gates in a circuit -is called T-count. In this work, we have focused not only on building the circuit using Clifford+T gates, but also on reducing the T-count to make the circuit as efficient as possible.
The proposed circuit will be built using only four gates: the CNOT gate, the Toffoli gate, the Temporary logical-AND gate, and the uncomputation gate of the Temporary logical-AND operation. Of this list of gates, only the CNOT gate is a direct member of the Clifford+T group. However, since the Clifford+T group is universal, the remaining gates can be approximated using only gates from this group. The most efficient implementation of the Toffoli gate in terms of T-count is the one proposed by Amy et al. [19] (Fig. 1a). It has a T-count of 7. However, the Temporary logical-AND gate, proposed by Gidney [11], has a T-count of only 4 (Fig. 1b). The Temporary logical-AND gate is similar to the Toffoli gate, but with a disadvantage: it can only act on a target qubit prepared in the state ( 1 ) . In other words, the Toffoli gate can be applied on any qubit, while the Temporary logical-AND gate can only be applied on a pre-prepared auxiliary qubit. Another important difference between the two gates is that to reverse a Toffoli operation, another Toffoli gate must be applied, thus doubling the T-count. Nevertheless, the uncomputation gate of the Temporary logical-AND gate can be used at no extra cost in terms of T-count thanks to the "measure and fixup" approach shown in Fig. 1c [11].

State-of-the-art in quantum circuit design for arithmetic operations
The importance of quantum circuits that implement arithmetic operations is due to the need for these operations in larger circuits and quantum algorithms of proven efficiency. An example already mentioned is Shor's algorithm, which requires adders for its implementation. It is precisely because of this algorithm that addition is the arithmetic operation that receives more interest from the scientific community [20]. As a consequence, there are a huge amount of adders for quantum computing published in the literature [5,20,21]. In general, such adders can be classified into two main types: ripple-carry adders, and carry-ahead adders. The former are characterized by requiring fewer resources than the latter, while the latter are faster at the cost of requiring more resources. Carry-ahead adders use these extra resources to calculate the carry bits before the sum, thus reducing the total time of the operation. For instance, in the case of adding two N bit numbers, such an adder will calculate the n-th carry (with n < N ) before the partial n − 1-th sum finishes. Therefore, carries can be added to these partial sums without waiting for the previous partial sum to finish. Apart from their methodology, the designs are usually focused on optimizing one or more metrics: the quantum cost, the T-count, etc. Likewise, there are designs focused on achieving implementations that meet certain circumstantial conditions, such as only Clifford+T type gates are used in their construction [5,21]. Although addition is the operation that receives most interest from the research community, the rest of the arithmetic operations are also necessary for various quantum algorithms. In the case of multiplication, there are several different works in the literature that have progressively optimized resources in terms of quantum gates, noise tolerance, or delay cost [22][23][24][25][26][27]. In the case of division, there are currently also several published works that, as with multiplication, have progressively optimized the use of the scarce resources currently available for quantum computers Symbol and implementation of (a) the Toffoli gate proposed by Amy et al. [19], (b) the Temporary logical-AND gate proposed by Gidney [11], and (c) the uncomputation gate of the Temporary logical-AND gate 1 3 [6,[28][29][30][31][32]. The need for these two operations is justified by the fact that they are involved in quantum algorithms focused on maximization of objective functions, quantum image processing, or the evaluation of transcendental functions [33][34][35]. Subtraction is also represented in the literature [36][37][38]. However, it is often performed using adders and two's complement representation (as in classical digital electronics) [20]. Given the aforementioned high interest that adders receive in quantum computing, which implies a high availability of this type of circuits, it is not surprising that the second option (using adders and two's complement representation to perform the subtraction) is the most widespread one [20]. Closely related to the latter are circuits called converters, which allow converting a signed binary number into two's complement. This operation can be performed using adders or subtractors. However, using a circuit specifically designed for this operation reduces costs and improves speed and noise tolerance [39][40][41].
Regarding the operation that is the focus of this paper, the computation of the absolute-value of a number, there are no circuits available in the literature at the time of writing or at least the authors have not been able to find any. The absolutevalue is involved, as already mentioned, in existing quantum circuits and algorithms. However, in such works, conditional adders and certain extra operations are used to perform the computation of the absolute value. This methodology is explained in detail in the next subsection. In this work, a circuit specifically dedicated to this operation is proposed to make it more efficient compared to how it is currently performed in the mentioned works.

Methodology to compute the absolute-value of a number
The absolute-value of a two's complement number can be obtained by reversing its sign or leaving it as it is if the number is negative or positive, respectively. This process is called taking the two's complement, and it is widely used in classical digital design [42]. This process begins by identifying the sign of the number, contained in its most-significant digit. If this digit is 0, the number is positive. Otherwise, the number is negative. It can be observed in Table 1 that the absolute value of a number is equal to the number itself in the case of positive numbers. Then, no operations are necessary if the number is positive. However, actions are required in the case of negative numbers. For such cases, all digits of the number must be inverted, and number 1 must be added to the value obtained after the inversions. For instance, Table 1 shows that the value −5 is represented as 1011. By inverting its digits, the value 0100 is obtained. Adding 0100 + 1 produces the result 0101, which corresponds to the value 5, the absolute value of −5.
It is also possible to obtain the absolute value of negative numbers represented in two's complement using a simple trick. This trick consists of going through the digits of the number starting from the least-significant one, that is, from right to left. Once the first 1 is found, the successive digits (following the direction of displacement to the left) must be inverted, leaving the first 1 and the 0 to its right (if any) invariant. Following the previous example, the absolute-value of 1011 ( −5 ) will be obtained by finding the least significant digit at 1, which in this case coincides with 1 3 Efficient design of a quantum absolute-value circuit using… the least significant digit of the number. Next, the digits to the left of that digit are inverted, obtaining the value 0101. Despite being a valuable trick for people, this second methodology is not used in digital electronics because it involves a greater number of operations to separate the subset of bits to be inverted from the bits that must remain invariant. [42,43]. Therefore we have focused on using the first methodology, which involves inverting all the digits using the sign of the number itself, as well as performing an addition. The first operation can be easily performed using CNOT gates, while the second can be computed using adder circuits, which are the most abundant arithmetic circuits in quantum computing [5,20,21].

Proposed implementation
In this section we present our proposed quantum absolute-value circuit. The circuit has no garbage outputs, and it involves less qubits and T gates than the circuits in the state-of-the-art. The circuit computes the absolute-value of a integer number B represented using two's complement. The digits of B are stored in a register of qubits �B N−1 ⟩...�B 1 ⟩�B 0 ⟩ , being N the number of digits of B. Extra qubits are involved in the operation. At the end of the circuit, the qubits that initially contain B will contain | B | with the exception of the one that contains the sign ( B N−1 , it maintains its value). The extra qubits, initialized to the specific states that will be detailed in this section, are unchanged after the computation of the circuit. The proposed circuit is based on the methodology of the conditional adder presented by Coreas and Thapliyal [26]. Figure 2 shows an example of the conditional adder. The circuit of Coreas and Thapliyal performs the addition of two numbers A and B if the control qubit (Ctrl in Fig. 2) is set to 1. Otherwise, the circuit just returns the original numbers A and B. This conditional adder, with some minor changes, can be used to compute the absolute-value of a number. It has been explained that the absolute-value of a number B can be obtained keeping B unchanged if it is positive, or flipping its digits and adding 1 to it otherwise. If the digit that contains the sign of B is used as control qubit, that is, Ctrl = B N−1 , and A is set to 1, the circuit will compute the addition B + 1 if B < 0 . However, B is required to previously flip its digits if it is negative. Such flips can be performed using N − 1 CNOT gates, setting B N−1 as the control qubit and B i (0 ≤ i < N − 1) as the target ones. The adaptation of the conditional adder to perform the absolute-value of a number B is shown in Fig. 3, for the N = 7 case.
The circuit shown in Fig. 3 is fully functional. However, it is possible to subject it to several optimizations and to obtain a more efficient circuit in terms of operations, T-count, and required number of qubits. In the circuit shown in Fig. 3, there are several qubits with constant values (those that contain A = 1 ). Some operations can be simplified as the qubits that contain values A j (∀j > 0) are set to 0. At the beginning of the conditional addition, there are N − 2 CNOT gates performing the operations 0 ⊕ B i (0 < i < N − 1) , being 0 the control qubit and B i the target one. The result of such operations will be always B i , so these CNOT gates can be removed. Those qubits, the ones that contains the 0 values acting as control qubits, also perform the operations A i ⊕ A i+1 (0 < i < N − 2) . These operations are now irrelevant, so the N − 3 involved CNOT gates can be removed. All these CNOT operations are uncomputed at the end of the circuit, so the CNOT gates dedicated to the uncomputation can also be removed from the circuit. Then, a total of 4N − 10 CNOT gates can be saved. On the other hand, several extra simplifications can be applied since A 0 is always 1. This qubit only acts as a target qubit in two Toffoli gates. Such Toffoli gates can be replaced by two CNOT gates. Moreover, the qubit A 0 is no longer needed. Taking into account all the simplifications listed above, the resulting circuit reduces the number of Toffoli gates by 2, the number of CNOT gates by 4N − 8 , and requires one less qubit. An example of this circuit, for the N = 7 case, is shown in Fig. 4. Minor changes can be applied over the circuit shown in Fig. 4. The first ancilla qubit is always initialized to 0 ( A 1 = 0 ). This qubit acts as a target qubit of the operation B 0 ⊕ 0 . After the operation, the target qubit will contain B 0 , and the qubit will be used as one of the control qubits in a Toffoli operation. This qubit is not involved in other gate until the uncomputation of the mentioned operations. The operation B 0 ⊕ 0 (that it, the CNOT gate) can be removed, and the qubit that initially contains B 0 can be used in the Toffoli gate as control qubit. Therefore, the  [26] (Fig. 3), setting A = 1 and being B the number whose absolutevalue is to be obtained. The most-significant digit of B, which contains its sign, will be the control qubit that decides if the addition is computed. Previously to the conditional addition, the digits of B should be flipped if the number is negative. This can be easily performed using the control qubit and N − 1 CNOT gates. In short, the circuit does nothing if B is positive ( B 6 = 0 ), and inverts the digits of B and adds 1 to it if B is negative. Here, S will contain the absolute-value of B. We keep the notation of the original circuit for the sake of clarity  Fig. 5. For the sake of clarity, this figure (and the previous ones) shows the gaps left by the gates and qubits that have been removed. A cleaner, more compact version of the circuit is shown in Fig. 6.
In the circuit of Fig. 6, there are N − 2 Toffoli gates simulating a (N − 1)-qubit Toffoli gate. Later, the same number of Toffoli gates are applied to uncompute the previous ones. Such gates acts over target qubits initialized to 0 (the uncomputation gates revert them into 0). These Toffoli gates can be replaced by Temporary logical-AND gates (and their corresponding uncomputation gates) without increasing the number of necessary qubits but halving the T-count of the circuit. Efficient design of a quantum absolute-value circuit using… the absolute-value now has a target qubit that is initialized to 0. Therefore, this Toffoli gate can be replaced by a Temporary logical-AND (saving a T-count of 3). The remaining Toffoli gates cannot be replaced as they act over qubits that contain dependent values. An example of the final circuit, for the N = 7 case, is shown in Fig. 7.
The described evolution can be summarised as follows: • Original version (Fig. 2): the circuit proposed by Muñoz-Coreas and Thapliyal. • Version 1 (Fig. 3): the original version, simply making A = 1 and being B the two's complement number whose absolute value is to be calculated. It is necessary to extract the most significant digit from this number, which contains the sign, and to use it as a control qubit of the addition (Ctrl). • Version 2 (Fig. 4): to eliminate qubit A 0 , knowing that it is always set to �1⟩ .
It only acts as a control qubit in some Toffoli gates, so these Toffoli gates can become CNOT gates dependent on the other control qubit. • Version 3 ( Fig. 5 and 6): to remove qubit A 1 knowing that it is always set to �0⟩ .
This qubit acts as the target qubit of a CNOT gate, and then acts as the control qubit in a Toffoli gate. The control qubit of the CNOT gate replaces A 1 as the control qubit in the Toffoli gate. • Version 4 ( Fig. 7): Qubits initialized to �0⟩ are replaced by qubits initialized in the state ( 1 ) . Toffoli gates that had an auxiliary qubit in state �0⟩ as a target qubit are replaced by temporary logical-AND gates. Toffoli gates that reverted to the replaced Toffoli gates are in turn replaced by uncomputation gates of the temporary logical-AND gate. 4. A Temporary logical-AND is applied to perform the operation B 0 B 1 . The result of the operation is temporarily stored in an ancilla qubit a 0 . 5. For i = 2 to N − 2 , a temporary logical-AND is applied to perform the operation a i−2 B i . The result of the operation is temporarily stored in an ancilla qubit a i−1 . 6. A Temporary logical-AND gate is used to compute B N−1 a N−3 . The result is saved in an ancilla qubit a N−2 , and it is the most significant digit of the absolute-value (excluding the sign, which is always assumed to be positive). 7. For i = N − 2 to 2, an uncomputation gate of the Temporary logical-AND operation is applied to uncompute the operation a i−2 B i , that is, to uncompute a i−1 . Then, a Toffoli gate must be applied to compute a i−2 B N−1 ⊕ B i . The target qubit will contain the i-digit of the absolute-value. 8. An uncomputation gate of the Temporary logical-AND operation is applied to revert the product B 0 B 1 (that is, a 0 ). Then, a Toffoli gate computing B 6 B 0 ⊕ B 1 , and a CNOT gate performing B 6 ⊕ B 0 , are applied to compute the least-significant digits of the result.

Results
The number of necessary gates to build the proposed circuit can be easily checked by going through the steps described in Sect. 3.1: • No gates are needed in steps 1 and 2. • Step 3 involves N − 1 CNOT gates. • Step 4 needs one Temporary logical-AND gate. • Step 5 computes N − 3 Temporary logical-AND gates. • Step 6 involves one Temporary logical-AND gate. • Step 7 computes N − 3 uncomputation gates of the Temporary logical-AND gate. N − 3 Toffoli gates are also applied. • Step 8 involves one uncomputation gate of the Temporary logical-AND operation. A Toffoli gate and a CNOT gate are also applied.
Overall, the proposed circuit needs N CNOT gates, N − 2 Toffoli gates, N − 1 Temporary logical-AND gates, and N − 2 uncomputation gates of the Temporary logical-AND operation. The original circuit, the conditional-adder of Muñoz-Coreas and Thapliyal [26], involves 5N − 11 CNOT gates, and 3N − 1 Toffoli gates (extra costs are added to compute the absolute-value operation using this circuit, as it is shown in Fig. 3). The proposed circuit reduces the number of CNOT gates in 4N − 11 with respect to the original circuit, and the number of Toffoli gates in 2N + 1 . The saved Toffoli gates are replaced by N − 1 extra Temporary logical-AND gates, and N − 2 uncomputation gates (of the Temporary logical-AND gate). This will have a very positive impact on the T-count, as explained below.
The T-count of the proposed circuit can be obtained from the number of Toffoli and Temporary logical-AND gates it involves. The remaining gates (CNOT and uncomputation gates) do not contain any T gates. Therefore, the circuit has a T-count of 11N − 18 . On the other hand, it is trivially obtained through steps 1 and

3
Efficient design of a quantum absolute-value circuit using… 2 of Sect. 3.1 that the circuit involves 2N − 1 qubits. Table 2 shows a comparison in terms of T-count and number of qubits between the proposed circuit and the best conditional-adders (acting as absolute-value circuits) in the literature. Since there are no specific circuits in the literature for the absolute value of two's complement integers, the comparison only includes such conditional-adders. Our proposal halves the T-count with respect the best circuit in the state-of-the-art: from 21N − 7 to 11N − 18 . Given the high cost of the T gate, this reduction represents a significant saving. In terms of qubits, our proposal also manages to reduce their number. It needs two fewer qubits than the circuit of Muñoz-Coreas and Thapliyal [26], four fewer qubits than the circuits of Markov and Saeedi [44] and Jayashree et al [45], and half as many qubits as circuit of Lin et al. [46].

Conclusions
We have presented a circuit to perform the absolute-value of signed integer numbers (in two's complement representation) of any size in quantum computers. The proposed circuit is optimized in terms of number of operations, T-count, and number of qubits. The circuit is based in the methodology of a conditional-adder. Its design has been described in detail, showing each necessary step, quantum gate, and qubit. It has been built using exclusively Clifford+T gates.
Finally, a comparison has been made in terms of T-count and number of qubits between the proposed circuit and the most important circuits available in the literature. In this comparison, it can be seen that our circuit improves on the others in terms of number of qubits and, what is more, halves the T-count with respect to these circuits.