Content addressable memory (CAM) is one of the functional memories which realize word-parallel equivalence search. Since a CAM unit is generally used in a particular application program, we consider that appropriate design for CAM units is required depending on the requirements for the application program. This paper proposes a hardware/software cosynthesis system for CAM processors. The input of the system is an application program written in C including CAM functions and a constraint for execution time (or CAM processor area). Its output is hardware descriptions of a synthesized processor and a binary code executed on it. Based on the branch-and-bound method, the system determines which CAM function is realized by a hardware and which CAM function is realized by a software with meeting the given timing constraint (or area constraint) and minimizing the CAM processor area (or execution time of the application program). We expect that we can realize optimal CAM processor design for an application program. Experimental results for several application programs show that we can obtain a CAM processor whose area is minimum with meeting the given timing constraint.

This paper proposes an area/time optimizing algorithm in a high-level synthesis system for control-based hardwares. Given a call graph whose node corresponds to a control flow of an application program, the algorithm generates a set of state-transition graphs which represents the input call graph under area and timing constraint. In the algorithm, first state-transition graphs which satisfy only timing constraint are generated and second they are transformed so that they can satisfy area constraint. Since the algorithm is directly applied to control-flow graphs, it can deal with control flows such as bit-wise processes and conditional branches. Further, the algorithm synthesizes more than one hardware architecture candidates from a single call graph for an application program. Designers of an application program can select several good hardware architectures among candidates depending on multiple design criteria. Experimental results for several control-based hardwares demonstrate effectiveness and efficiency of the algorithm.

This paper proposes a hardware/software cosynthesis system for digital signal processor cores and a hardware/software partitioning algorithm which is one of the key issues for the system. The target processor has a VLIW-type core which can be composed of a processor kernel, multiple data memory buses (X-bus and Y-bus), hardware loop units, addressing units, and multiple functional units. The processor kernel includes five pipeline stages (RISC-type kernel) or three pipeline stages (DSP-type kernel). Given an application program written in the C language and a set of application data, the system synthesizes a processor core by selecting an appropriate kernel (RISC-type or DSP-type kernel) and required hardware units according to the application program/data and the hardware costs. The system also generates the object code for the application program and a software environment (compiler and simulator) for the processor core. The experimental results demonstrate that the system synthesizes processor cores effectively according to the features of an application program and the synthesized processor cores execute most application programs with the minimum number of clock cycles compared with several existing processors.

Recently, high-level synthesis techniques for FPGA designs (FPGA-HLS techniques) are strongly required in various applications. Both interconnection delays and clock skews have a large impact on circuit performance implemented onto FPGA, which indicates the need for floorplan-driven FPGA-HLS algorithms considering them. To appropriately estimate interconnection delays and clock skews at HLS phase, a reasonable model to estimate them becomes essential. In this paper, we demonstrate several experiments to characterize interconnection delays and clock skews in FPGA and propose novel estimate models called “IDEF” and “CSEF”. In order to evaluate our models, we integrate them into a conventional floorplan-driven FPGA-HLS algorithm. Experimental results demonstrate that our algorithm can realize FPGA designs which reduce the latency by up to 22% compared with conventional approaches.

As device feature size decreases, interconnection delay becomes the dominating factor of circuit total delay. Distributed-register architectures can reduce the influence of interconnection delay. They may, however, increase circuit area because they require many local registers. Moreover original distributed-register architectures do not consider control signal delay, which may be the bottleneck in a circuit. In this paper, we propose a high-level synthesis method targeting generalized distributed-register architecture in which we introduce shared/local registers and global/local controllers. Our method is based on iterative improvement of scheduling/binding and floorplanning. First, we prepare shared-register groups with global controllers, each of which corresponds to a single functional unit. As iterations proceed, we use local registers and local controllers for functional units on a critical path. Shared-register groups physically located close to each other are merged into a single group. Accordingly, global controllers are merged. Finally, our method obtains a generalized distributed-register architecture where its scheduling/binding as well as floorplanning are simultaneously optimized. Experimental results show that the area is decreased by 4.7% while maintaining the performance of the circuit equal with that using original distributed-register architectures.

This paper focuses on SIMD processor synthesis and proposes a SIMD instruction set/functional unit synthesis algorithm. Given an initial assembly code and a timing constraint, the proposed algorithm synthesizes an area-optimized processor core with optimal SIMD functional units. It also synthesizes a SIMD instruction set. The input initial assembly code is assumed to run on a full-resource SIMD processor (virtual processor) which has all the possible SIMD functional units. In our algorithm, we introduce the SIMD operation decomposition and apply it to the initial assembly code and the full-resource SIMD processor. By gradually reducing SIMD operations or decomposing SIMD operations, we can finally find a processor core with small area under the given timing constraint. The promising experimental results are also shown.

Approximate computing is a promising solution for future energy-efficient designs because it can provide great improvements in performance, area and/or energy consumption over traditional exact-computing designs for non-critical error-tolerant applications. However, the most challenging issue in designing approximate circuits is how to guarantee the pre-specified computation accuracy while achieving energy reduction and performance improvement. To address this problem, this paper starts from the state-of-the-art general approximate adder model (GeAr) and extends it for more possible approximate design candidates by relaxing the design restrictions. And then a maximum-error-distance-based performance/accuracy formulation, which can be used to select the performance/energy-accuracy optimal design from the extended design space, is proposed. Our evaluation results show the effectiveness of the proposed method in terms of area overhead, performance, energy consumption, and computation accuracy.

Due to the increase of outsourcing by IC vendors, we face a serious risk that malicious third-party vendors insert hardware Trojans very easily into their IC products. However, detecting hardware Trojans is very difficult because today's ICs are huge and complex. In this paper, we propose a hardware-Trojan classification method for gate-level netlists to identify hardware-Trojan infected nets (or Trojan nets) using a support vector machine (SVM) or a neural network (NN). At first, we extract the five hardware-Trojan features from each net in a netlist. These feature values are complicated so that we cannot give the simple and fixed threshold values to them. Hence we secondly represent them to be a five-dimensional vector and learn them by using SVM or NN. Finally, we can successfully classify all the nets in an unknown netlist into Trojan ones and normal ones based on the learned classifiers. We have applied our machine-learning-based hardware-Trojan classification method to Trust-HUB benchmarks. The results demonstrate that our method increases the true positive rate compared to the existing state-of-the-art results in most of the cases. In some cases, our method can achieve the true positive rate of 100%, which shows that all the Trojan nets in an unknown netlist are completely detected by our method.

With the progress of process technology in recent years, low voltage power supplies have become quite predominant. With this, the voltage margin has decreased and therefore the on-chip decoupling capacitance optimization that satisfies the voltage drop constraint becomes more important. In addition, the reduction of the on-chip decoupling capacitance area will reduce the chip area and, therefore, manufacturing costs. Hence, we propose an algorithm that satisfies the voltage drop constraint and at the same time, minimizes the total on-chip decoupling capacitance area. The proposed algorithm uses the idea of the network algorithm where the path which has the most influence on voltage drop is found. Voltage drop is improved by adding the on-chip capacitance to the node on the path. The proposed algorithm is efficient and effectively adds the on-chip capacitance to the greatest influence on the voltage drop. Experimental results demonstrate that, with the proposed algorithm, real size power/ground network could be optimized in just a few minutes which are quite practical. Compared with the conventional algorithm, we confirmed that the total on-chip decoupling capacitance area of the power/ground network was reducible by about 4050%.

This paper proposes a hardware/software cosynthesis algorithm for processors with heterogeneous registers. Given a CDFG corresponding to an application program and a timing constraint, the algorithm generates a processor configuration minimizing area of the processor and an assembly code on the processor. First, the algorithm configures a datapath which can execute several DFG nodes with data dependency at one cycle. The datapath can execute the application program at the least number of cycles. The branch and bound algorithm is applied and all the number of functional units and memory banks are tried. For an assumed number of functional units and memory banks, an appropriate number of heterogeneous registers and connections to functional units and registers are explored. The experimental results show effectiveness and efficiency of the algorithm.

This paper presents a test input data compression technique, Selective Low-Care Coding (SLC), which can be used to significantly reduce input test data volume as well as the external test channel requirement for multiscan-based designs. In the proposed SLC scheme, we explored the linear dependencies of the internal scan chains, and instead of encoding all the specified bits in test cubes, only a smaller amount of specified bits are selected for encoding, thus greater compression can be expected. Experiments on the larger benchmark circuits show drastic reduction in test data volume with corresponding savings on test application time can be indeed achieved even for the well-compacted test set.

A scan chain is one of the most important testing techniques, but it can be used as side-channel attacks against a cryptography LSI. We focus on scan-based attacks, in which scan chains are targeted for side-channel attacks. The conventional scan-based attacks only consider the scan chain composed of only the registers in a cryptography circuit. However, a cryptography LSI usually uses many circuits such as memories, micro processors and other circuits. This means that the conventional attacks cannot be applied to the practical scan chain composed of various types of registers. In this paper, a scan-based attack which enables to decipher the secret key in an AES cryptography LSI composed of an AES circuit and other circuits is proposed. By focusing on bit pattern of the specific register and monitoring its change, our scan-based attack eliminates the influence of registers included in other circuits than AES. Our attack does not depend on scan chain architecture, and it can decipher practical AES cryptography LSIs.

Power analysis for IoT devices is strongly required to protect attacks from malicious attackers. It is also very important to reduce power consumption itself of IoT devices. In this paper, we propose a highly-adaptable and small-sized in-field power analyzer for low-power IoT devices. The proposed power analyzer has the following advantages: (A) The proposed power analyzer realizes signal-averaging noise reduction with synchronization signal lines and thus it can reduce wide frequency range of noises; (B) The proposed power analyzer partitions a long-term power analysis process into several analysis segments and measures voltages and currents of each analysis segment by using small amount of data memories. By combining these analysis segments, we can obtain long-term analysis results; (C) The proposed power analyzer has two amplifiers that amplify current signals adaptively depending on their magnitude. Hence maximum readable current can be increased with keeping minimum readable current small enough. Since all of (A), (B) and (C) do not require complicated mechanisms nor circuits, the proposed power analyzer is implemented on just a 2.5cm×3.3cm board, which is the smallest size among the other existing power analyzers for IoT devices. We have measured power and energy consumption of the AES encryption process on the IoT device and demonstrated that the proposed power analyzer has only up to 1.17% measurement errors compared to a high-precision oscilloscope.

Non-volatile memories are paid attention to as a promising alternative to memory design. Data stored in them still may be destructed due to crosstalk and radiation. We can restore the data by using error-correcting codes which require extra bits to correct bit errors. Further, non-volatile memories consume ten to hundred times more energy than normal memories in bit-writing. When we configure them using error-correcting codes, it is quite necessary to reduce writing bits. In this paper, we propose a method to generate a bit-write-reducing code with error-correcting ability. We first pick up an error-correcting code which can correct t-bit errors. We cluster its codeswords and generate a cluster graph satisfying the S-bit flip conditions. We assign a data to be written to each cluster. In other words, we generate one-to-many mapping from each data to the codewords in the cluster. We prove that, if the cluster graph is a complete graph, every data in a memory cell can be re-written into another data by flipping at most S bits keeping error-correcting ability to t bits. We further propose an efficient method to cluster error-correcting codewords. Experimental results show that the bit-write-reducing and error-correcting codes generated by our proposed method efficiently reduce energy consumption. This paper proposes the world-first theoretically near-optimal bit-write-reducing code with error-correcting ability based on the efficient coding theories.

A packed SIMD type operation or a SIMD operation is n-parallel b/n-bit sub-operations executed by the modified n-bit functional unit. Such a functional unit is called a SIMD functional unit and a processor core which can execute SIMD operations is called a SIMD processor core. SIMD operations can be effectively applied to image processing applications. This paper focuses on hardware/software cosynthesis of SIMD processor cores and particularly proposes a new simulator generator which simulates pipelined instructions for a SIMD processor. Generally, a SIMD functional unit has many options and then we can have so many different SIMD functional unit instances. However, since our hardware/software cosynthesis system synthesizes a special-purpose processor core for an input application program, it uses very limited SIMD functional unit instances. In the proposed approach, we consider a SIMD operation to be a set of SIMD sub-operations. By adding up the appropriate SIMD sub-operations, we construct a single SIMD operation. Then a SIMD functional unit behavior can be characterized by a collection of SIMD operations. This approach has the advantage that: if we have a small number of behavior libraries for SIMD sub-operations, we can instantiate a particular SIMD functional unit behavior. Experimental results demonstrate the effectiveness of the proposed approach.

Recently, it has been reported that malicious third-party IC vendors often insert hardware Trojans into their products. Especially in IC design step, malicious third-party vendors can easily insert hardware Trojans in their products and thus we have to detect them efficiently. In this paper, we propose a machine-learning-based hardware-Trojan detection method for gate-level netlists using multi-layer neural networks. First, we extract 11 Trojan-net feature values for each net in a netlist. After that, we classify the nets in an unknown netlist into a set of Trojan nets and that of normal nets using multi-layer neural networks. By experimentally optimizing the structure of multi-layer neural networks, we can obtain an average of 84.8% true positive rate and an average of 70.1% true negative rate while we can obtain 100% true positive rate in some of the benchmarks, which outperforms the existing methods in most of the cases.

CAM (Content Addressable Memory) units are generally designed so that they can be applied to variety of application programs. However, if a particular application runs on CAM units, some functions in CAM units may be often used and other functions may never be used. We consider that appropriate design for CAM units is required depending on the requirements for a given application program. This paper proposes a CAM processor synthesis system based on behavioral descriptions. The input of the system is an application program written in C including CAM functions, and its output is hardware descriptions of a synthesized processor and a binary code executed on it. Since the system determines functions in CAM units and synthesizes a CAM processor depending on the requirements of an application program, we expect that a synthesized CAM processor can execute the application program with small processor area and delay. Experimental results demonstrate its efficiency and effectiveness.