DESIGN AND IMPLEMENTATION OF LOW AREA AND LOW POWER AES ENCRYPTION HARWARE CORE

By Group No.16 1. VAIBHAV GUPTA (0809131251) 2. VIJAY KUMAR VERMA (0809131095) 3. SHIVANI CHAURASIA (0809131082)

Under the guidance of Mr.SAMPATH KUMAR DEPARTMENT OF ELECTRONICS & COMMUNICATION ENGINEERING

JSS ACADEMY OF TECHNICAL EDUCATION C-20/1 SECTOR-62, NOIDA

April, 2012

Chapter 1 INTRODUCTION

In todays digital world, encryption is emerging as a disintegrable part of all communication networks and information processing systems, for protecting both stored and in transit data. Encryption is the transformation of plain data (known as plaintext) into unintelligible data (known as ciphertext) through an algorithm referred to as cipher. There are numerous encryption algorithms that are now commonly used in computation, but the U.S. government has adopted the Advanced Encryption Standard (AES) to be used by Federal departments and agencies for protecting sensitive information. The National Institute of Standards and Technology (NIST) has published the specifications of this encryption standard in the Federal Information Processing Standards (FIPS) Publication 197.

Any conventional symmetric cipher, such as AES, requires a single key for both encryption and decryption, which is independent of the plaintext and the cipher itself. It should be impractical to retrieve the plaintext solely based on the ciphertext and the encryption algorithm, without knowing the encryption key. Thus, the secrecy of the encryption key is of high importance in symmetric ciphers such as AES. Software implementation of encryption algorithms does not provide ultimate secrecy of the key since the operating system, on which the encryption software runs, is always vulnerable to attacks.

There are other important drawbacks in software implementation of any encryption algorithm, including lack of CPU instructions operating on very large operands, word size mismatch on different operating systems and less parallelism in software. In addition, software implementation does not fulfill the required speed for time critical encryption applications. Thus, hardware implementation of encryption algorithms is an important alternative, since it provides ultimate secrecy of the encryption key, faster speed and more efficiency through higher levels of parallelism.

Different versions of AES algorithm exist today (AES128, AES196, and AES256) depending on the size of the encryption key. In this project, a hardware model for implementing the AES128 algorithm was developed using the SystemVerilog hardware description language. A unique feature of the design proposed in this project is that the round keys, which are consumed during different iterations of encryption, are generated in parallel with the encryption process.

The hardware model was then completely verified using a testbench, which took advantage of the SystemVerilogs object oriented programming (OOP) feature, by constructing random test objects and providing them to the model. The validation process continued until the model was verified for a certain Functional Coverage. Then, the verified model was synthesized using the Synopsis Design-Compiler tool to get an estimate of the number of gates, area and timing of the hardware model.

In addition, the AES128 algorithm was modeled in C language and was ported on a Simics virtual system. The statistics of the Simics virtual system was gathered to get an estimate of the time it would take to encrypt a plaintext block on the virtual system. Finally, the performances of software and hardware implementations were compared.

Chapter 2

ADVANCED ENCRYPTION STANDARD (AES)

2.1 OVERVIEW
This chapter is a summary of the Federal Information Processing Standards (FIPS) Publication 197 , issued by the National Institute of Standards and Technology (NIST) which specifies the Advanced Encryption Standard. Throughout the remainder of this chapter, the mathematical properties of the Advanced Encryption Standard (AES) are introduced using the information obtained from the AES specification. The AES is a subset of a much larger encryption algorithm known as Rijndael, which was one of many proposals to the NIST competing for becoming a standard encryption algorithm. On October of 2000, the NIST announced the Rijndael algorithm as the winner due to the best overall score in security, performance, efficiency, implementation capability and simplicity. The AES algor ithm is a symmetric cipher. In symmetric ciphers, a single secret key is used for both the encryption and decryption, whereas in asymmetric ciphers, there are two sets of keys known as private and public keys. The plaintext is encrypted using the public key and can only be decrypted using the private key. In addition, the AES algorithm is a block cipher as it operates on fixed-length groups of bits (blocks), whereas in stream ciphers, the plaintext bits are encrypted one at a time, and the set of transformations applied to successive bits may vary during the encryption process. The AES algorithm operates on blocks of 128 bits, by using cipher keys with lengths of 128, 192 or 256 bits for the encryption process. Although the original Rijndael encryption algorithm was capable of processing different blocks sizes as well as using several other cipher key lengths, but the NIST did not adopt these additional features in the AES.

2.2 INPUTS, OUTPUTS AND THE STATE

The plaintext input and ciphertext output for the AES algorithms are blocks of 128 bits. The cipher key input is a sequence of 128, 192 or 256 bits. In other words the length of the cipher key, Nk, is either 4, 6 or 8 words which represent the number of columns in the cipher key. The AES algorithm is categorized into three versions based on the cipher key length. The number of rounds of encryption for each AES version depends on the cipher key size. In the AES algorithm, the number of rounds is represented by Nr, where Nr = 10 when Nk = 4, Nr = 12 when Nk = 6, and Nr = 14 when Nk = 8. The following table illustrated the variations of the AES algorithm. For the AES algorithm the block size (Nb), which represents the number of columns comprising the State is Nb = 4.

TABLE 2.1-AES VARIATIONS

The basic processing unit for the AES algorithm is a byte. As a result, the plaintext, ciphertext and the cipher key are arranged and processed as arrays of bytes. For an input, an output or a cipher key denoted by a, the bytes in the resulting array are referenced as an , where n is in one of the following ranges: Block length = 128 bits, 0 <= n < 16 Key length = 128 bits, 0 <= n < 16 Key length = 192 bits, 0 <= n < 24 Key length = 256 bits, 0 <= n < 24

All byte values in the AES algorithm are presented as the concatenation of their individual bit values between braces in the order {b7, b6, b5, b4, b3, b2, b1, b0}. These

bytes are interpreted as finite field elements using a polynomial representation:

As an example, {10001001} (or {85} in hexadecimal) identifies the polynomial .The arrays of bytes in the AES algorithm are represented as .

All the AES algorithm operations are performed on a two dimensional 4x4 array of bytes which is called the State, and any individual byte within the State is referred to as Sr,c, where letter r represent the row and letter c denotes the column. At the beginning of the encryption process, the State is populated with the plaintext. Then the cipher performs a set of substitutions and permutations on the State. After the cipher operations are conducted on the State, the final value of the state is copied to the cipher text output as is shown in the following figure.

Figure 2.2 State Population and Results

At the beginning of the cipher, the input array is copied into the State according the following scheme: s[r,c] = in [r + 4c] for 0 <= r < 4 and 0 <= c < 4 , and at the end of the cipher the State is copied into the output array as shown below: out[r+4c] = s[r,c] for 0 <= r < 4 and 0 <= c < 4.

2.3 Cipher Transformations

The AES cipher either operates on individual bytes of the State or an entire row/column. At the start of the cipher, the input is copied into the State as described in Section 2.2. Then, an initial Round Key addition is performed on the State. Round keys are derived from the cipher key using the Key Expansion routine. The key expansion routine generates a series of round keys for each round of transformations that are performed on the State. The transformations performed on the state are similar among all AES versions but the number of transformation rounds depends on the cipher key length. The final round in all AES versions differs slightly from the first Nr 1 rounds as it has one less transformation performed on the State. Each round of AES cipher (except the last one) consists of all the following transformation: - SubBytes( ) - ShiftRows( ) - MixColumns( ) - AddRoundKey ( ) The AES cipher is described as a pseudo code in Figure 2.3. As shown in the pseudo code, all the Nr rounds are identical with the exception of the final round which does not include the MixColumns transformation. The array w[] represents the round keys that are generated by the key expansion routine. In the following sections, individual transformations that are used in each encryption round are described.

Figure 2.3 AES Cipher

2.3.1 - Subbytes ( ) transformation The SubBytes is a byte substitution operation performed on individual bytes of the State, as shown in Figure 2.4, using a substitution table called S-box.

Figure 2.4 SubBytes Transformation

The invertible S-box table is constructed by performing the following transformation on each byte of the State. - Take the multiplicative inverse in the finite field GF(28) of the byte. - Apply the following transformation to the byte:

The bi is the ith bit of the byte and ci is the ith bit of a constant byte with the value of {63}. The combination of the two transformations can be expressed in matrix form as shown below:

The S-box table shown in Table 2 is constructed by performing the two transformations described earlier for all possible values of a byte, ranging from {00} to {ff}. For example the substitution value for {53} would be determined by the intersection of the row with index 5 and the column with index 3.

Table 2.5 AES S-box

2.3.2 - Shiftrows ( ) transformation

The ShiftRows transformation cyclically shifts the last three rows of the state by different offsets. The first row is left unchanged in this transformation. Each byte of the second row is shifted one position to the left. The third and fourth rows are shifted left by two and three positions, respectively. The ShiftRows transformation is illustrated in Figure 2.6.

Figure 2.6 ShiftRows Transformation

2.3.3 Mixcolumns ( ) transformation This transformation operates on the columns of the State, treating each columns as a four term polynomial the finite field GF(2^8). Each columns is multiplied modulo x4+1 with a fixed four-term polynomial a(x) = {03}x3 + {01}x2 + {01}x + {02} over the GF(2^8). The MixColumns transformation can be expressed as a matrix multiplication as shown below:

The MixColumns transformation replaces the four bytes of the processed column with the following values:

The corresponds to the multiplication of polynomials in GF(2^8) modulo an irreducible polynomial of degree 8. A polynomial is irreducible if its only divisors are one and itself. For the AES algorithm the irreducible polynomial is:

The MixColumns transformation is illustrated in Figure 2.7. This transformation together with ShiftRows, provide substantial diffusion in the cipher meaning that the result of the cipher depends on the cipher inputs in a very complex way. In other words, in a cipher with a good diffusion, a single bit change in the plaintext will completely change the ciphertext in an unpredictable manner.

Figure 2.7 MixColumns Transformation

2.3.4 AddRoundKey ( ) Transformation

During the AddRoundKey transformation, the round key values are added to the State by means of a simple Exclusive Or (XOR) operation. Each round key consists of Nb words that are generated from the KeyExpansion routine. The round key values are added to the columns of the state in the following way:

In the equation above, the round value is between r 0 round N . When round=0, the cipher key itself is used as the round key and it corresponds to the initial AddRoundKey transformation displayed in the pseudo code in Figure 2.3. The AddRoundKey transformation is illustrated in Figure 2.8.

Figure 2.8 AddRoundKey Transformation

2.4 AES KEY EXPANSION

The AES algorithm requires four words of round keys for each encryption round. That is total of 4*(Nr + 1) round keys considering the initial set of keys required for the first AddRoundKey transformation. All the round keys are derived from the cipher key itself.

According to the Federal Information Processing Standards (FIPS) Publication 197 , there is no restriction on the cipher key selection, as no week cipher key has been identified for the AES algorithm. The expansion of the cipher key into the round keys is performed by the KeyExpansion algorithm as shown in the pseudo code in Figure 2.9.

Figure 2.9 KeyExpansion Algorithm

In the above pseudo code, the array w[] represents the round keys that are generated by the KeyExpansion routine and Nk represents the size of the cipher key. Depending on the version of the AES algorithm, Nk=4, 6 or 8. The first Nk words of the expanded key are filled with the cipher key. The SubWord( ) function applies the same S-box substitution to each of the four bytes in the word. The RotWord( ) function takes a word [a0,a1,a2,a3] as input and perform a cyclic shift and returns the word [a1,a2,a3,a0]. The round constant word array, Rcon[i], contains a 32 bit value given by [{02}i-1,{00},{00},{00}]. Every following round key , w[i], is equal to the XOR of the previous round key, w[i-1], and the word Nk positions earlier, w[i-Nk]. For words in positions that are a multiple of Nk, two transformations are initially applied to the previous round key, w[i-1]. These transformations are a cyclic shift of the bytes in the previous round key, followed by the application of the S-box table lookup to all four bytes of the word. Afterwards, an XOR with a round constant value, Rcon[i], is applied to the previous round key. The KeyExpansion routine for the AES256 (Nk=8) is slightly different than the AES128 and AES192 ones, as an additional SubWord function is applied to the previous round key, w[i-1], prior to the XOR with w[i- Nk].

CHAPTER 3 AES128 DESIGN AND IMPLEMENTATION

3.1 OVERVIEW
In this chapter, a hardware model for implementing the AES128 algorithm is introduced. The model is implemented using the SystemVerilog hardware description language . This chapter covers the design and implementation issues of the AES128 algorithm. In the next chapter, a test infrastructure is presented that thoroughly tests the functionality of the implemented model. The hardware model developed in this chapter is synthesizable. This means that the model provides a cycle-by-cycle RTL description of the circuit that a logic synthesis tool can convert to an optimized gate-level netlist. The modeling process utilized in this project is the bottom-up approach. This means that the leaf components in the design hierarchy were developed first and the higher-level modules were constructed by instantiating their subcomponents and connecting them with the internal signals. All the modules in the design hierarchy were modeled in behavioral style, but the root module consisted of data flow modeling as well to implement the four major cipher transformations.

3.2 DESIGN HIERARCHY

The proposed AES128 hardware model is a 3-level hierarchical design as shown in Figure 3.1. The root module in the hierarchy is the AES128_cipher_top. This module implements the AES128 pseudo code displayed in Figure 2.3. It has two 128-bit inputs for receiving the cipher key and the plaintext. There is also a single bit input signal, Ld, which is used to indicate the availability of a new set of plaintext or cipher key on the input ports. The completion of the encryption process is indicated by asserting the done single bit output.

Figure 3.1 Design Hierarchy

A unique feature of the proposed design is that the AES128_Key_Expand module is pipelined with the AES128_cipher_top module. While the AES128_cipher_top module is performing an iteration of the encryption transformations on the State using the previously generated round keys, the AES128_Key_Expand produces the next rounds set of keys to be used by the root module in the next encryption iteration.

3.2.1 Aes128 Encryption Process

The AES128_cipher_top module state diagram is shown in Figure 3.2. There are ten rounds of transformations represented by r1 to r10 states. The four cipher transformations introduced in section 2.3 are applied to each state. The r0 state corresponds to the initial AddRoundKey transformation in Figure 2.3.

After leaving the Reset state, the AES128_Cipher_Top module waits for assertion of the Ld signal, which indicates that a valid set of plaintext and cipher key is available on the input ports. After reaching the r0 state, there is a transition on every clock cycle for the next ten cycles, as ten rounds of encryption is applied to the State. After going through ten rounds of transformations, the done signal is asserted to indicate the completion of cipher and availability of the ciphertext on the corresponding output port.

Figure 3.2 AES128_Cipher_Top Module State Diagram

3.2.2 AES128 Round Key Generation The round keys used by the AES128_Cipher_Top module are generated based on the state diagram shown in Figure 3.3. The AES128_Key_Expand and the AES128_RCon modules are responsible for generating the round keys. These two modules operate based on the state diagram shown in Figure 3.3, which is slightly different than the one used for the encryption process.

Figure 3.3 AES128_Key_Expand Module State Diagram

In the state diagram shown above, the Ld signal is checked in the r0 state and if asserted, then the cipher key is provided to the AES128_Cipher_Top module to be used for the initial AddRoundKey transformation.

The AES128_Key_Expand module generates four 32-bit keys for each round of the encryption process, by using the cipher key. Figure 12 shows the block diagram of the AES128_Key_Expand module. The cipher key is passed to this module through a 128-bit input port, and the round keys are generated on the four output ports.

Figure 3.4 AES128_Key_Expand Module

There is a 32-bit round constant value, which is used by the key expansion algorithm to generate the round keys. This value varies for each encryption round and for Nr=1 to Nr=10 is given by [{02}i-1,{00},{00},{00}]. The AES128_RCcon module is used to generate this value as shown in Figure 13. The AES128_RCon module also operates based on the state diagram shown in Figure 10.

Figure 3.5 AES128_Rcon Module

3.3 AES128 PIPELINED DESIGN

As stated earlier in this chapter, the round key generation in the proposed design is pipelined with the encryption rounds. The pipelined operation of the round key expansion and the cipher is shown in Figure 11. Each AES encryption round n (white cells) is pipelined with the key generation for round n+1 (gray cells).

Figure 3.6 AES128 Pipelined Round Key Generation and Cipher Rounds

The most important advantage of the pipelined design is the lower delay for each encryption iteration, since the round keys for each encryption iteration is present at the beginning of the iteration cycle. The lower delay in each encryption iteration means faster completion of each round of encryption. This reduces the overall encryption delay and allows the design to operate at higher clock frequencies. The higher clock frequency will increase the message encryption rate (throughput) making this design suitable for time critical encryption applications.

CHAPTER 4 AES128 VERIFICATION

4.1 OVERVIEW
In this chapter, we describe the test infrastructure that is developed in SystemVerilog to verify the functionality of the model described in the previous chapter. The simulation was done using the Synopsis VCS tool. The testbench fully validated the design by constructing random cyclic test vectors for the plaintext and the cipher key, passing them to the model, and comparing the ciphertext to the expected result.

4.2 Testbench Infrastructure

There are four major steps involved in verifying a design using an HDL, including test vector generation, passing the test vectors to the design and capturing the design response, determining correctness by comparing the design response with the expected results, and measuring the verification coverage. The test infrastructure described in this chapter performs all the above steps in a systematic way. The AES128 test infrastructure contains several components, some of which are unique SystemVerilog features. These SystemVerilog features make the verification of a design more reliable and more structured. The test infrastructure components are displayed in Figure 4.1 as part of the AES128_Top module.

Figure 4.1 AES128 Test Infrastructure

The test infrastructure utilizes the SystemVerilog program block, which has multiple implicit timing regions to evaluate the design events separately from the testbench events. The program block is connected to the model through another unique feature of the SystemVerilog, called Interface. The Interface bundles the connections between the testbench and the design while enforcing the synchronization and communication protocol between the two entities. The definition of the AES128_Top module in SystemVerilog is shown in Figure 4.2, which has the high-level instantiation of the modules constructing the test infrastructure.

Figure4.2 AES128_Top Definition

The AES128_Top module instantiates the design, Interface and the Program. The Interface and the Program constructs are discussed in the next two sections. The clock generator is defined inside the AES128_top module as well, to avoid any potential race conditions.

4.3 Aes128_Interface
As designs are becoming more complex, the number of module ports and the complexity of the interconnections between the modules are also increasing. The SystemVerilog Interface construct is the solution for properly connecting the modules as it provides an intelligent means of communication between several modules. The Interface bundles the ports together and enforces synchronization between the modules connected through it. The Interface can provide connectivity between design modules and/or testbench. The modport construct is used in an Interface to specify the direction of signals that are bundled together and to group the signals that are synchronous to a specifc clock. In this project, the SystemVerilog Interface was only used to connect the high-level design with the testbench as shown in Figure 4.1. As a result, there were two modports declared for the Interface in this project. In an Interface, the signals that are synchronous to a clock are defined inside a Clocking Block to ensure correct timing between the testbench and the high-level design. This ensures that any synchronous signal is driven or sampled with respect to clock and eliminates the potential race condition that exists between the testbench and high-level design written in Verilog. The AES128_Interface definition is shown in Figure 4.3.

Figure 4.3 AES128_Interface Definition

4.3 AES128_Program
In Verilog, a testbench is basically another module which is connected to the highlevel design. This can cause a race condition between the testbench and the design. System Verilog hardware description language introduces a new construct called Program to be used as the testbench. The SystemVerilog Program, having one (or more entry) points, is closer to a program in C, than Verilogs many small blocks of concurrently executing hardware . It also has multiple implicit timing regions to evaluate the design events separately from the testbench event, eliminating any race condition between the design under test and the testbench. The testbench described in this chapter consists of a single Program, which uses the Object Oriented Programming feature of SystemVerilog to dynamically build random test vectors. This is done by defining a Class inside the AES128_Program that encapsulates two random cyclic variables (Properties) for generating stimulus to the high-level design. The class defined in the AES128_Program is shown in Figure 4.4.

As stated earlier in this chapter, another important feature of a testbench is keeping track of the verification coverage. In other words, to make sure that a design is thoroughly verified, the testbench needs to test all the design features. Functional Coverage is a measure of which design features have been exercised by the test. Functional Coverage is done by means of Cover Groups defined inside the System Verilog Program. Each Cover Group consists of multiple Cover Points that are the variables used for generating stimulus for the design under test. As it is shown in Figure 4.4, the class defined in the AES128_Program uses a single Cover Group to keep track of the 128-bit plain_text and cipher_key stimuli. Due to limitations of the Synopsys VCS compiler that limits the cyclic random objects to no more than 16 bits, the 128-bit stimuli are broken into arrays of 16-bit elements. Each array element is declared as a Cover Point inside the Cover Group to be sampled together for measuring the Functional Coverage.

Figure 4.4 Class Definition in the AES128_Program

The AES128_Program pseudo code is shown in Figure 4.5. This testbench verifies the design until the Functional Coverage is 100%. The verification procedure involves generating the stimuli and passing them through the AES128_Interface to the design under test and verifying correctness of the results obtained from the design.

Figure 4.5 AES128_Program Pseudo Code

To verify the correct functionality of the design under test, a C-style function is developed in SystemVerilog, which takes the stimuli as input and calculates the expected ciphertext. This function is defined as part of package that contains all the variables and routines involved in the encryption process as shown in Figure 4.6.

Figure 4.6 AES128_Testbench_Package pseudo code

The complete simulation result of the testbench is included in Appendix C. Figure 4.7 illustrates the simulation result for the first three test cases. Each test case starts with randomizing the cover points to populate the plaintext and cipher key inputs to the design under test. Then, the expected ciphertext is calculated using the AES128_cipher function shown in Figure 4.6. After the design under test has encrypted the plaintext and the done signal is asserted, the ciphertext generated by the hardware model is compared with the expected result to catch any mismatch. The last step in each test case is gathering the Functional Coverage and continuing with the next test case until all design features are tested.

Figure 4.7 Sample Simulation Results

CHAPTER 5 AES128 SYNTHESIS

5.1 Overview
A primary objective of this project was to develop a synthesizable model for the AES128 encryption algorithm. Synthesis is the process of converting the register transfer level (RTL) representation of a design into an optimized gate-level netlist. This is a major step in ASIC design flow that takes an RTL model closer to a low-level hardware implementation. Synthesis consists of three main steps. The first step is the Translation, which involves converting the RTL description of a design into a non-optimized intermediate representation that is used by the synthesis tool. The second step is the logic optimization, which optimizes the internal representation by removing redundant logic and performing Boolean logic optimizations. The third step is called technology mapping & optimization which maps the internal representation to an optimized gate level representation using the technology library cells based on design constraints.[3] In this chapter, we describe how the Synopsys Design_Compiler tool was utilized to synthesize the verified AES128 model, by using a script that was developed to perform the synthesis based on certain constraints. The script generates several reports about the synthesis outcome including timing and area estimates.

5.2 Synthesis Methodology

The first step in the synthesis process is to read all the components in the design hierarchy. There are three components in the 3-level design hierarchy that needs to be synthesized. Since the RTL model utilizes a SystemVerilog Package, then the synthesis tool needs to enable the semantics of a package. In addition, the synthesis tool needs to know if there are multiple instances of calling an automatic function in the design, to preserve separate values for each instance.

The following Synopsys Design Compiler (DC) shell commands enable package and automatic function utilizations: set hdlin_sv_packages "enable" set hdlin_infer_function_local_latches "true" Then, the package and the modules in the design hierarchy are read using the following commands: read_file read_file read_file read_file -format -format -format -format sverilog sverilog sverilog sverilog {./AES128_DUT_package.sv} {./AES128_rcon.sv} {./AES128_key_expand.sv} {./AES128_cipher_top.sv}

After reading the design files, they are Analyzed and Elaborated through which the RTL code is converted into the Synopsys Design Compiler internal format. The intermediate results are stored in the defined working library. The following DC commands are used for these steps: analyze -library WORK -format sverilog {./AES128_rcon.sv} analyze -library WORK -format sverilog {./AES128_key_expand.sv} analyze -library WORK -format sverilog {./AES128_cipher_top.sv} elaborate AES128_rcon -architecture verilog -library WORK elaborate AES128_key_expand -architecture verilog -library WORK elaborate AES128_cipher_top -architecture verilog -library WORK Then, the dont_touch attribute is removed from all the modules in the design hierarchy so that during the optimization phase the tool can modify the modules. The following DC command is used for this step: remove_attribute [find design -hierarchy] dont_touch After this step, a 40MHz clock signal is applied to the clock port of the root module, and the synthesis tool is programmed not to modify the clock tree during the optimization phase. In addition, an arbitrary input delay of 5ns with respect to the clock port is applied to all input and output ports (except the clock port itself) to set a safe margin by considering any unintended source of delay such as the delay associated with driving module/modules. Then, the design is constrained with hypothetical maximum area equal to zero to force the tool to make the gate level netlist as compact as possible. The following DC commands are used for these steps:

create_clock -name clk -period 25 [find port intf_clk] set_dont_touch_network [find clock "clk"] set non_clock_ports [remove_from_collection [all_inputs] [get_ports intf_clk]] set_input_delay 5 $non_clock_ports -clock clk set_output_delay 5 [all_outputs] set_max_area 0 In the next steps, the tool is programmed to consider a unique design for each cell instance by removing the multiply-instantiated hierarchy in the current design. Then, the synthesis script removes the boundaries from all the components in the design hierarchy and removes all levels of hierarchy. uniquify set_boundary_optimization [find design -hierarchy] true ungroup -all -flatten -all_instances Finally, the tool compiles the design with high effort and reports any warning related the mapping and final optimization step. At the end, the tool generates reports for the optimized gate level netlist area, the worst combinational path timing, and any violated design constraint. report_attribute > ./Synthesis_Reports_Attribute.txt report_area > ./Synthesis_Reports_Area.txt report_constraints -all_violators > ./Synthesis_Reports_Constraint_Violaters.txt report_timing -path full -delay max -max_paths 1 -nworst 1 > ./Synthesis_Reports_Timing.txt

5.3 Synthesis Timing Result

The synthesis tool optimizes the combinational paths in a design. In General, four types of combinational paths can exist in any design: 1- Input port of the design under test to input of one internal flip-flip 2- Output of an internal flip-flip to input of another flip-flip 3- Output of an internal flip-flip to output port of the design under test

4- A combinational path connecting the input and output ports of the design under test. The last DC command in the script developed in previous section, instructs the tool to report the path with the worst timing. In this case, the path with the worst timing is a combinational path of type two. The delay associated with this path is the summation of delays of all combinational gates in the path plus the Clock-To-Q delay of the originating flip-flop, which was calculated as 24.09ns. By considering the setup time of the destination flip-flop in this path, which is 0.85ns, the 40MHz clock signal satisfies the worst combinational path delay. The delays of combinational gates, setup time of flip41 flops and Clock-To-Q values are derived from the LSI_10k library file that was used for the mapping step during synthesis. The synthesis timing report is shown below:
Report : timing -path full -delay max -max_paths 1 Design : AES128_cipher_top Version: Z-2007.03 Date : Mon Nov 16 21:25:14 2009 Operating Conditions: Wire Load Model Mode: top Startpoint: u0/w3_reg[22] (rising edge-triggered flip-flop clocked by clk) Endpoint: u0/w2_reg[27] (rising edge-triggered flip-flop clocked by clk) Path Group: clk Path Type: max

5.4 Synthesis Area Result

The synthesis area report shows the total number of cells and nets in the netlist. It also uses the area parameter associated with each cell in the LSI_10K library file, to calculate the total combinational and sequential area of the netlist. The total area of the gate level netlist is unknown since it depends on total area of the interconnects, which itself is a function of the wiring load model used in physical design. The total cell area in the netlist is reported as 22978 units, which is the sum of combinational and sequential areas. The synthesis area report is shown below:

5.5 Synthesis Constraint Violators Result

To enforce the synthesis tool to create the most compact netlist, the area of the gate level netlist was constrained to zero during the synthesis process. As a result, the only constraint violation, which is expected, is related to the area as shown below:

CHAPTER 6 AES128 SOFTWARE IMPLEMENTATION

6.1 Overview
The optimized gate level netlist generated after synthesizing the hardware model by using the LSI_10K technology library can operate at a 40MHz clock signal. Since the hardware model takes ten clock cycles (for ten rounds of encryption) to encrypt a 128-bit block, the overall delay for encrypting a block of plaintext is 250ns. In order to compare the speed of the hardware implementation with that of a software implementation, the AES128 algorithm was modeled in C language. The C program was then run on a virtual system, and the statistics of the virtual system were gathered before and after encrypting a block of plaintext. The number of CPU cycles that were required on the virtual system to encrypt a block of plaintext was used to compare the efficiency of software and hardware implementations.

6.2 AES128 Software Implementation on a Simics Virtual System

Simics is a complete functional simulation tool for creating virtual platforms that supports single-core, multicore, multiple processor, and multiple machine configurations (racks, clusters, and distributed systems). Simics supports several processor families (e.g. ARM, MIPS, PowerPC, x86) and runs the same binary software as the physical target system. To the target software, the virtualized target hardware behaves exactly the same as the physical target hardware. [8] In this project, the Simics software was used to create a virtual system based on Intels x86 architecture and the 440BX chipset. The target virtual system consisted of a 2GHz Pentium4 processor and ran the Red Hat 7.3 Enterprise Linux operating system.

The C program implementing the AES128 encryption algorithm (See Appendix

D) was ported to the Simicss virtual system and then compiled to create the executable file (object file). The virtual systems statistics were gathered during the execution of the C program, before and after encrypting a block of plaintext. This was done by using the Simicss Magic instruction that called a registered python function for gathering the virtual system statistics. The portion of the C code for encrypting a block of plaintext is shown in Figure 4.7. Encrypting a block of plaintext involves copying the block to the state, generating the round keys from the cipher key and performing ten rounds of encryption on the state.

Figure 6.1 AES128 Block Encryption Pseudo Code in C

The target virtual system statistics before and after encryption of a plaintext block is summarized in Table 6.2. The Callback1 and Callback2 statistics refer to the virtual systems state before and after the encryption of the plaintext block, respectively.

Table 6.2 Simics Virtual System Statistics

The User and Supervisor columns refer to the number of instruction that were executed in the user space and the system space, respectively. Since the clock per instruction for the virtual target was assumed to be one (CPI=1), the total CPU cycles was equal to the total number of instructions. The results show that encrypting a block of plaintext in software takes more than 30,000 CPU cycles of the virtual target system. Since the virtual system has a 2GHz Pentium4 processor, the encryption of a plaintext block takes more than 15us, which is 60 times slower than the proposed hardware implementation.