rus
The article proposes the structure and principles of implementing a photon computer. At the heart of its functioning are the effects of interaction of coherent systems of light waves generated with a laser source. Estimates of productivity values, energy consumption, physical dimensions are obtained; these estimates show the possible advantages of a photonic computer in comparison with an electronic computer.
Теги: computer technology optical logic elements photon computer photon processor вычислительная техника оптические логические элементы фотонный компьютер фотонный процессор
INTRODUCTION
To solve constantly emerging problems, computational machines with maximum performance are required. It is measured by the number of operations performed per time unit.
Modern computers contain quadrillions of electronic transistors measuring no more than 14 nm, perform ~1017 op/s, consume dozens of MW, occupy thousands of m2.
A further reduction in the size of the transistors for increasing the capacity leads to the fact that the interaction between them will be accompanied by quantum effects. Computers based on such effects are called quantum computers. They have been studied for the past 30 years, the prospects for their implementation are still uncertain [1].
Therefore, the proposed below is a computer that is based on the interaction effects of coherent systems of light waves generated by laser radiation [2]. The carriers of radiation are photons; hence its name is photonic computer.
The photon computer, unlike the analogues [3, 4], operates only with the light representation of information. This allows to obtain, as will be shown, the values of productivity and energy efficiency better than those achieved using electronic technologies. This article is an expanded version of the publication [5].
ALGORITHM OF FUNCTIONING AND STRUCTURE OF PHOTONIC COMPUTER
The program recorded in a high-level language is translated by an electronic computer into an electronic form of a photonic computer program, which is then converted into a light form and executed by a photonic processor. At the end of the calculation, the result is transformed by the coupling device into an electronic form and enters the computer.
The structure of the photonic computer is shown in Fig. 1. The execution of the computational process begins with the starting action of the device 1, transmitted via channel 2 to the laser source 3. The radiation emitted by the source 3 passes through the optical channel 4 to the input/output device 5, where it is separated into light beams, the number of which is equal to the number of bits simultaneously input through the optical channels 6 into the processing elements (PE) 7, to the photonic processor 9. Further, as a result of the interaction of these beams in the device 5 with the electronic representation of the information input via the electronic channel 10 from the device 1 (which can be used as a computer), a light form of the program is obtained, coming via channels 6 for processing into PE7. They contain arithmetic logic units (ALU), switches (SW), and control devices (CD), combined by optical channels 11. The computing process is a sequence of light beam interactions in passive optical logical elements (OLE) [6, 7, 8], of which PE7 is implemented. The OLE functions are identical to the functions of electronic logic units [9]. This allows you to use architectural implementations of ALU, SW and CD in the photonic computer, known from electronic computer technology; the use of optical delay lines is possible for synchronization.
The processor elements in the photonic processor are connected by channels 8 to a multiprocessor environment having any known topology, for example, a 3D-torus, a Gn-hypercube, etc. [10].
The total number of operations performed is determined either by a power budget [11] or by a coherence length [12].
After the exhaustion of the power budget, the information is regenerated. After reaching the coherence length, the information is returned through the channels 6 to the device 5, where it is converted into an electronic form and then again into a light coherent form transmitted to the PE.
PRINCIPLES OF COMPUTATION
The information in the light form constantly moves in space. To exclude its delays and power losses, it is necessary:
• to perform operations for the readiness of operands without access to memory according to the control discipline from the data flow [9];
• to process information on conflict-free routes [10], using only free processor elements and channels at a certain, known time interval.
The computed expression is represented by the reverse polish notation [9], on the basis of which the multilevel structure (MLS) of the algorithm is constructed [13]; the vertices of one level of the MLS correspond to the commands performed independently (in parallel) during the same interval t.
A processor graph is obtained from the algorithm MLS. Its vertices correspond to a PE executing operations at time intervals t. They are connected by edges belonging to non-conflicting routes from channels 8. Through these channels, PEs exchange their identifiers, i. e. operation codes, operands and synchronizing symbols, which are denoted by Ш. To perform the operations, PEs that are free in interval t from other tasks are assigned; these PEs are determined when preparing the program in device 1 based on the number of PE, the topology of their connection, belonging to the conflict-free routes.
The program is obtained from the processor graph by replacing the vertices with photonic computer commands.
During the calculation each identifier is accompanied by the current value t = 0, 1, …, m, where m is the depth of the algorithm MLS [13]. The value of t is incremented by one for each "passing" of the PE.
The processor elements that are not used in the execution of program instructions in each interval receive and transmit the symbol Ш and the value of t (increasing t by one).
The processor element performs the operation only when the data with the same successive value of t is received through all its channels.
EXAMPLE OF THE COMPUTING PROCESS
Let’s plot a processor graph to calculate the value of the expression A = a + (b + c) Ч d by a photonic computer, the PEs of which is combined in G3 – three-dimensional hypercube [10] as shown in Fig. 2. The numbers of the notations in Fig. 1 and Fig. 2 coincide.
Each vertex of G3 means the processor element to which the vector corresponds , where, σi = 1,2,3.
The edges connecting the vertices of G3 correspond to the channels 8, they are labeled 81, 82, 83.
Furthermore, PEs are connected to the device 5 by channels 61, 62, 63, external for G3. Fig. 2 shows channels 61, 62, 63 only for PE at the vertex (0,0,1).
For G3 and the route length 3, we have [10] four conflict-free sets ((000), (111)), (001), (110)), ((010), (101)) and ((100), (011)). Their vertices and the conflict-free routes connecting these peaks are marked in Fig. 2 by black, green, red and blue, respectively. The directions of the transfer are shown by arrows.
Tables 1 and 2 for the elements (0,0,0) and (0,0,1) give vectors corresponding to the processor elements, from which through the channels 81, 82, 83 to the intervals t = 0,1, …, 6 by conflict-free routes receive information.
The algorithm MLS for the expression A = a + (b + c) Ч d in the record A a b c + d Ч + := is shown in Fig. 3, the processor graph is shown in Fig. It consists of the vertices corresponding to the processor elements (σ3, σ2, σ1), in time intervals t and denoted (σ3, σ2, σ1) t. The vertices are connected by edges from the channels 8 belonging to the conflict-free routes.
The identifiers are loaded respectively into the vertices (0,0,1)0 – the identifier of the operation {Ч} and the values of "b" and "d", in (0,1,0)0 – the values of "A", "a" and "c", in (1,0,0)0 – the identifiers of the operation {+}, {+} and {:=}. The symbols listed follow from device 5 through channels 6 as shown in Fig. 4. The remaining elements are followed by the symbols Ш.
This is shown, in particular, in Fig. 4 the route of the identifier A in the intervals t = 0,1,2,3,4 consists of the vertices (0,1,0)0, (1,1,0)1, (0,1,0)2, (0, 1.1)3, (0.0.1)4. The characters routes are shown in partial, with dotted lines.
By replacing the vertices of the graph with commands, we obtain a program for a photonic computer. The program consists of the commands that prescribe the actions to be performed at each interval t by each PE, for each channel.
We distinguish three phases of the command:
• phase ϕ = 1; reception of information from channels to PE;
• phase ϕ = 2; execution of PE actions prescribed by the operation code;
• phase ϕ = 3; transfer of information from PE to receivers in other PEs.
Table 3 shows the photonic processor program with the G3 topology, obtained from the processor graph shown in Fig. In each column (σ3, σ2, σ1) for ϕ = 1, the notation means that the identifier x is taken over the channel 8i PE in the interval t = z (σ3, σ2, σ1); ϕ = 2 indicates that the operation is performed by the PE data during the interval t = z, the dash indicates no operation; for ϕ = 3, the record means that in the interval z the identifier x is transmitted via the 8i PE channel (σ3, σ2, σ1); the color of the line in the record is the same as the color of the route on which the information is received or transmitted.
The details of the translation of programs written in high-level languages and the features of obtaining the "machine" representation of these programs are inessential here. All known means can be used to optimize the translation and the generated machine code.
FEATURES OF PHOTONIC PROCESSOR IMPLEMENTATION
Photonic processor communications topologies
The photonic processor is a multiprocessor environment consisting of PEs connected by communication channels.
It is possible to use any known [10] topology of the connections – 1D torus, 2D torus, 3D torus, Nm, Gn-hypercube, N-full matrix switch, S-star, F-fully connected tree realized from q-port switches. The criteria for selection are the values of Dk – the number of routes between two PEs located at a distance k (measured by the minimum number of edges between them) indicated in Table 4, as well as the values of C and G in Table 5 – the power and the number of conflict-free sets that are estimated relative to environment diameter – the minimum value of k, sufficient to connect any two elements of the environment containing ω PEs. The values of Dk, C, and G were obtained in [10]. They determine the communication capabilities of the environment and, consequently, the parameters of the processor graphs and programs.
The "primitive" environment is 1D torus. The best values of C and G have the most complex N and S environments in the hardware implementation of the environment. The environments with topologies F and Gn are acceptable in practice for hardware complexity. These environments are characterized by close values of Dk, C, and G.
The environment of Gn has a convenient arrangement for constructing conflict-free routes and a formal description of Boolean functions, it provides for topological backup facilities that automatically replace failed PE with full maintenance of the environment topology and computational process [10].
Interaction with the environment
The photonic processor’s exchange with the external environment should take as little time as possible. This is achieved if each PE photonic processor has access to the external environment, for example, is assigned with an electronic processor that implements the functions of storing and exchanging information.
The electronic processors corresponding to the elements of the photonic processor can be combined by communication channels into a multiprocessor environment, in particular, having the same topology as the photonic processor. This allows you to use the same description of the electronic and photonic components of the photonic computer.
Element base of the photonic processor
Element base is a functionally complete set of OLE that allows performing any arithmetic and logical operations. Such a set (non-unique) is the elements realizing the three logical functions "AND", "OR", "NOT" [9].
These elements can be realized by applying various effects, for example, interference [6], nonlinear [7], and others [8]. They are passive, they do not require additional pumping energy. All operations are carried out only with the light-form presentation of information. The duration of the operation is determined by the duration of light propagation in the OLE.
ESTIMATIONS OF THE PARAMETERS
We assume that the photonic computer shown in Fig. 2 involves a laser radiation source with a power PΣ = 100 W and a wavelength λ = 1530 nm; the power of radiation entering the input waveguides of the PE is P1 = 1 mW, the OLE minimum optical power at the input, based on the interference effect [6], is 100 nW [11].
The optical power budget of the PE will be b · 40 dB, where b = PΣ/8 · P1≈104 pcs. – the number of input OLE in each PE.
The efficiency of the radiation input into the waveguide is assumed to be 90% [14]. We assume that the OLE and waveguides losses coincide and are equal to Δ = 0.1–1.0 dB/cm [14]. The total length of the OLE waveguides of one PE does not exceed L = b · (0.35–3.5) m.
Assuming the length of the OLE waveguide D = 50 λ = 75 · 10–6 m, the diameter of the OLE input waveguide from [6] h = 7.5 · 10–6 m, we find [12] the distance between the OLE inputs l = 7.5 · 10–6 m.
One PE can house ρ = L/D = 4.7 · 107–4.7 · 108 pcs. of OLE.
The required amount of OLE in the n = 64 bit multiplier according to [9] does not exceed (23 n2 + 5 n – 58) ≈ 105 pcs.
Hardware-wise, the CD is not more complicated than the ALU [9], therefore, ρ = L/D = 4.7 · 107–4.7 · 108 pcs. of OLE are sufficient for implementation of PE containing 230–2300 ALU. It is known [9] that the duration of the operation of multiplication of n-bit numbers is t* ≈ 41.5 · n · τ, where τ is the operation time of the element, here τ = D/υ is the delay time in one OLE, υ = 2 · 108 m/s is the speed of light in the waveguide. For n = 64, the performance of the ALU is π = 109 op/s, the performance of the PE π* = 2.3 · 1011–2.3 · 1012 opt/s, the performance of the entire photon computer πΣ = 8 · π* ≈ 2 · 1012–2 · 1013 op/s.
Using architectural means to optimize the structure of ALU and PE [9], it is possible to obtain the result of the operation every interval τ and to reach the peak performance of a photonic computer . For λ = 1530 nm and D = 50 λ we have ≈ 5 · 1015 op/s – 5 · 1016 op/s.
The values of πΣ and for different λ when PΣ = 100 W are summarized in Table 6.
For OLE from [6], a decrease in the value of D by a factor of z (for example, as a result of decreasing λ) increases the processor’s performance in z2 times (with a constant power budget and equal losses).
The estimates of productivity and energy efficiency / PΣ of the photon computer is 103–104 times higher at λ = 1530 nm, D = 50 λ and Δ = 0.1–1.0 dB/cm than those achieved by modern computers [15].
To arrange PE, a silicon substrate with a length of no more than D · 4 is required, 7 · 108 = 352 · 102 m and width s ≈ 100 µm (obviously s > h and s > 2l). Area of the PE is 35 · 10–1 m2. With a substrate thickness of 100 µm, the volume of the eight PE will be V = 0.28 Ч 10–3 m3.
This is an acceptable size of the structural element that allows 100 W of heat to be drawn off by means of conventional air cooling.
CONCLUSION
The efficiency of information processing in the light form is achieved by joint application in a photonic computer of:
• passive optical logic elements
• computational discipline on the readiness of operands
• conflict-free algorithms for processing information by processor elements connected to a multiprocessor environment.
Classes of tasks solved by photonic and electronic computers coincide.
The completed analysis and the obtained estimates demonstrate the possibility of achieving a peak performance by a photonic computer using radiation with a wavelength of 1530 nm, 103–104 times greater than that achieved by modern electronic computing devices with equal energy consumed.
The use of technologies that ensure the use of ultraviolet radiation in a photonic computer with a wavelength of 100 nm allows the photonic computer to achieve Exaflop performance (1018 op/s) per 100 watts of power.
The radiation with different wavelengths do not interact with each other. If the OLE is properly implemented in one photonic computer, several computational processes can be performed simultaneously, represented by light waves of different lengths.
To solve constantly emerging problems, computational machines with maximum performance are required. It is measured by the number of operations performed per time unit.
Modern computers contain quadrillions of electronic transistors measuring no more than 14 nm, perform ~1017 op/s, consume dozens of MW, occupy thousands of m2.
A further reduction in the size of the transistors for increasing the capacity leads to the fact that the interaction between them will be accompanied by quantum effects. Computers based on such effects are called quantum computers. They have been studied for the past 30 years, the prospects for their implementation are still uncertain [1].
Therefore, the proposed below is a computer that is based on the interaction effects of coherent systems of light waves generated by laser radiation [2]. The carriers of radiation are photons; hence its name is photonic computer.
The photon computer, unlike the analogues [3, 4], operates only with the light representation of information. This allows to obtain, as will be shown, the values of productivity and energy efficiency better than those achieved using electronic technologies. This article is an expanded version of the publication [5].
ALGORITHM OF FUNCTIONING AND STRUCTURE OF PHOTONIC COMPUTER
The program recorded in a high-level language is translated by an electronic computer into an electronic form of a photonic computer program, which is then converted into a light form and executed by a photonic processor. At the end of the calculation, the result is transformed by the coupling device into an electronic form and enters the computer.
The structure of the photonic computer is shown in Fig. 1. The execution of the computational process begins with the starting action of the device 1, transmitted via channel 2 to the laser source 3. The radiation emitted by the source 3 passes through the optical channel 4 to the input/output device 5, where it is separated into light beams, the number of which is equal to the number of bits simultaneously input through the optical channels 6 into the processing elements (PE) 7, to the photonic processor 9. Further, as a result of the interaction of these beams in the device 5 with the electronic representation of the information input via the electronic channel 10 from the device 1 (which can be used as a computer), a light form of the program is obtained, coming via channels 6 for processing into PE7. They contain arithmetic logic units (ALU), switches (SW), and control devices (CD), combined by optical channels 11. The computing process is a sequence of light beam interactions in passive optical logical elements (OLE) [6, 7, 8], of which PE7 is implemented. The OLE functions are identical to the functions of electronic logic units [9]. This allows you to use architectural implementations of ALU, SW and CD in the photonic computer, known from electronic computer technology; the use of optical delay lines is possible for synchronization.
The processor elements in the photonic processor are connected by channels 8 to a multiprocessor environment having any known topology, for example, a 3D-torus, a Gn-hypercube, etc. [10].
The total number of operations performed is determined either by a power budget [11] or by a coherence length [12].
After the exhaustion of the power budget, the information is regenerated. After reaching the coherence length, the information is returned through the channels 6 to the device 5, where it is converted into an electronic form and then again into a light coherent form transmitted to the PE.
PRINCIPLES OF COMPUTATION
The information in the light form constantly moves in space. To exclude its delays and power losses, it is necessary:
• to perform operations for the readiness of operands without access to memory according to the control discipline from the data flow [9];
• to process information on conflict-free routes [10], using only free processor elements and channels at a certain, known time interval.
The computed expression is represented by the reverse polish notation [9], on the basis of which the multilevel structure (MLS) of the algorithm is constructed [13]; the vertices of one level of the MLS correspond to the commands performed independently (in parallel) during the same interval t.
A processor graph is obtained from the algorithm MLS. Its vertices correspond to a PE executing operations at time intervals t. They are connected by edges belonging to non-conflicting routes from channels 8. Through these channels, PEs exchange their identifiers, i. e. operation codes, operands and synchronizing symbols, which are denoted by Ш. To perform the operations, PEs that are free in interval t from other tasks are assigned; these PEs are determined when preparing the program in device 1 based on the number of PE, the topology of their connection, belonging to the conflict-free routes.
The program is obtained from the processor graph by replacing the vertices with photonic computer commands.
During the calculation each identifier is accompanied by the current value t = 0, 1, …, m, where m is the depth of the algorithm MLS [13]. The value of t is incremented by one for each "passing" of the PE.
The processor elements that are not used in the execution of program instructions in each interval receive and transmit the symbol Ш and the value of t (increasing t by one).
The processor element performs the operation only when the data with the same successive value of t is received through all its channels.
EXAMPLE OF THE COMPUTING PROCESS
Let’s plot a processor graph to calculate the value of the expression A = a + (b + c) Ч d by a photonic computer, the PEs of which is combined in G3 – three-dimensional hypercube [10] as shown in Fig. 2. The numbers of the notations in Fig. 1 and Fig. 2 coincide.
Each vertex of G3 means the processor element to which the vector corresponds , where, σi = 1,2,3.
The edges connecting the vertices of G3 correspond to the channels 8, they are labeled 81, 82, 83.
Furthermore, PEs are connected to the device 5 by channels 61, 62, 63, external for G3. Fig. 2 shows channels 61, 62, 63 only for PE at the vertex (0,0,1).
For G3 and the route length 3, we have [10] four conflict-free sets ((000), (111)), (001), (110)), ((010), (101)) and ((100), (011)). Their vertices and the conflict-free routes connecting these peaks are marked in Fig. 2 by black, green, red and blue, respectively. The directions of the transfer are shown by arrows.
Tables 1 and 2 for the elements (0,0,0) and (0,0,1) give vectors corresponding to the processor elements, from which through the channels 81, 82, 83 to the intervals t = 0,1, …, 6 by conflict-free routes receive information.
The algorithm MLS for the expression A = a + (b + c) Ч d in the record A a b c + d Ч + := is shown in Fig. 3, the processor graph is shown in Fig. It consists of the vertices corresponding to the processor elements (σ3, σ2, σ1), in time intervals t and denoted (σ3, σ2, σ1) t. The vertices are connected by edges from the channels 8 belonging to the conflict-free routes.
The identifiers are loaded respectively into the vertices (0,0,1)0 – the identifier of the operation {Ч} and the values of "b" and "d", in (0,1,0)0 – the values of "A", "a" and "c", in (1,0,0)0 – the identifiers of the operation {+}, {+} and {:=}. The symbols listed follow from device 5 through channels 6 as shown in Fig. 4. The remaining elements are followed by the symbols Ш.
This is shown, in particular, in Fig. 4 the route of the identifier A in the intervals t = 0,1,2,3,4 consists of the vertices (0,1,0)0, (1,1,0)1, (0,1,0)2, (0, 1.1)3, (0.0.1)4. The characters routes are shown in partial, with dotted lines.
By replacing the vertices of the graph with commands, we obtain a program for a photonic computer. The program consists of the commands that prescribe the actions to be performed at each interval t by each PE, for each channel.
We distinguish three phases of the command:
• phase ϕ = 1; reception of information from channels to PE;
• phase ϕ = 2; execution of PE actions prescribed by the operation code;
• phase ϕ = 3; transfer of information from PE to receivers in other PEs.
Table 3 shows the photonic processor program with the G3 topology, obtained from the processor graph shown in Fig. In each column (σ3, σ2, σ1) for ϕ = 1, the notation means that the identifier x is taken over the channel 8i PE in the interval t = z (σ3, σ2, σ1); ϕ = 2 indicates that the operation is performed by the PE data during the interval t = z, the dash indicates no operation; for ϕ = 3, the record means that in the interval z the identifier x is transmitted via the 8i PE channel (σ3, σ2, σ1); the color of the line in the record is the same as the color of the route on which the information is received or transmitted.
The details of the translation of programs written in high-level languages and the features of obtaining the "machine" representation of these programs are inessential here. All known means can be used to optimize the translation and the generated machine code.
FEATURES OF PHOTONIC PROCESSOR IMPLEMENTATION
Photonic processor communications topologies
The photonic processor is a multiprocessor environment consisting of PEs connected by communication channels.
It is possible to use any known [10] topology of the connections – 1D torus, 2D torus, 3D torus, Nm, Gn-hypercube, N-full matrix switch, S-star, F-fully connected tree realized from q-port switches. The criteria for selection are the values of Dk – the number of routes between two PEs located at a distance k (measured by the minimum number of edges between them) indicated in Table 4, as well as the values of C and G in Table 5 – the power and the number of conflict-free sets that are estimated relative to environment diameter – the minimum value of k, sufficient to connect any two elements of the environment containing ω PEs. The values of Dk, C, and G were obtained in [10]. They determine the communication capabilities of the environment and, consequently, the parameters of the processor graphs and programs.
The "primitive" environment is 1D torus. The best values of C and G have the most complex N and S environments in the hardware implementation of the environment. The environments with topologies F and Gn are acceptable in practice for hardware complexity. These environments are characterized by close values of Dk, C, and G.
The environment of Gn has a convenient arrangement for constructing conflict-free routes and a formal description of Boolean functions, it provides for topological backup facilities that automatically replace failed PE with full maintenance of the environment topology and computational process [10].
Interaction with the environment
The photonic processor’s exchange with the external environment should take as little time as possible. This is achieved if each PE photonic processor has access to the external environment, for example, is assigned with an electronic processor that implements the functions of storing and exchanging information.
The electronic processors corresponding to the elements of the photonic processor can be combined by communication channels into a multiprocessor environment, in particular, having the same topology as the photonic processor. This allows you to use the same description of the electronic and photonic components of the photonic computer.
Element base of the photonic processor
Element base is a functionally complete set of OLE that allows performing any arithmetic and logical operations. Such a set (non-unique) is the elements realizing the three logical functions "AND", "OR", "NOT" [9].
These elements can be realized by applying various effects, for example, interference [6], nonlinear [7], and others [8]. They are passive, they do not require additional pumping energy. All operations are carried out only with the light-form presentation of information. The duration of the operation is determined by the duration of light propagation in the OLE.
ESTIMATIONS OF THE PARAMETERS
We assume that the photonic computer shown in Fig. 2 involves a laser radiation source with a power PΣ = 100 W and a wavelength λ = 1530 nm; the power of radiation entering the input waveguides of the PE is P1 = 1 mW, the OLE minimum optical power at the input, based on the interference effect [6], is 100 nW [11].
The optical power budget of the PE will be b · 40 dB, where b = PΣ/8 · P1≈104 pcs. – the number of input OLE in each PE.
The efficiency of the radiation input into the waveguide is assumed to be 90% [14]. We assume that the OLE and waveguides losses coincide and are equal to Δ = 0.1–1.0 dB/cm [14]. The total length of the OLE waveguides of one PE does not exceed L = b · (0.35–3.5) m.
Assuming the length of the OLE waveguide D = 50 λ = 75 · 10–6 m, the diameter of the OLE input waveguide from [6] h = 7.5 · 10–6 m, we find [12] the distance between the OLE inputs l = 7.5 · 10–6 m.
One PE can house ρ = L/D = 4.7 · 107–4.7 · 108 pcs. of OLE.
The required amount of OLE in the n = 64 bit multiplier according to [9] does not exceed (23 n2 + 5 n – 58) ≈ 105 pcs.
Hardware-wise, the CD is not more complicated than the ALU [9], therefore, ρ = L/D = 4.7 · 107–4.7 · 108 pcs. of OLE are sufficient for implementation of PE containing 230–2300 ALU. It is known [9] that the duration of the operation of multiplication of n-bit numbers is t* ≈ 41.5 · n · τ, where τ is the operation time of the element, here τ = D/υ is the delay time in one OLE, υ = 2 · 108 m/s is the speed of light in the waveguide. For n = 64, the performance of the ALU is π = 109 op/s, the performance of the PE π* = 2.3 · 1011–2.3 · 1012 opt/s, the performance of the entire photon computer πΣ = 8 · π* ≈ 2 · 1012–2 · 1013 op/s.
Using architectural means to optimize the structure of ALU and PE [9], it is possible to obtain the result of the operation every interval τ and to reach the peak performance of a photonic computer . For λ = 1530 nm and D = 50 λ we have ≈ 5 · 1015 op/s – 5 · 1016 op/s.
The values of πΣ and for different λ when PΣ = 100 W are summarized in Table 6.
For OLE from [6], a decrease in the value of D by a factor of z (for example, as a result of decreasing λ) increases the processor’s performance in z2 times (with a constant power budget and equal losses).
The estimates of productivity and energy efficiency / PΣ of the photon computer is 103–104 times higher at λ = 1530 nm, D = 50 λ and Δ = 0.1–1.0 dB/cm than those achieved by modern computers [15].
To arrange PE, a silicon substrate with a length of no more than D · 4 is required, 7 · 108 = 352 · 102 m and width s ≈ 100 µm (obviously s > h and s > 2l). Area of the PE is 35 · 10–1 m2. With a substrate thickness of 100 µm, the volume of the eight PE will be V = 0.28 Ч 10–3 m3.
This is an acceptable size of the structural element that allows 100 W of heat to be drawn off by means of conventional air cooling.
CONCLUSION
The efficiency of information processing in the light form is achieved by joint application in a photonic computer of:
• passive optical logic elements
• computational discipline on the readiness of operands
• conflict-free algorithms for processing information by processor elements connected to a multiprocessor environment.
Classes of tasks solved by photonic and electronic computers coincide.
The completed analysis and the obtained estimates demonstrate the possibility of achieving a peak performance by a photonic computer using radiation with a wavelength of 1530 nm, 103–104 times greater than that achieved by modern electronic computing devices with equal energy consumed.
The use of technologies that ensure the use of ultraviolet radiation in a photonic computer with a wavelength of 100 nm allows the photonic computer to achieve Exaflop performance (1018 op/s) per 100 watts of power.
The radiation with different wavelengths do not interact with each other. If the OLE is properly implemented in one photonic computer, several computational processes can be performed simultaneously, represented by light waves of different lengths.
Readers feedback
