I. INTRODUCTION

Recently, a demand for IoT Devices will increase year by year, with 3.3 billions of equipment linked together by 2020 and the market size increase up to 457 Billion^[1,2]. In these IoT markets, network security and information protection are the most important factors in IoT device management. While existing IoT devices are distributed, the following security issues can occur due to problems that need to be controlled by the central server^[2,3].

1) Malicious attacks on networks and devices that could lead to system failure

2) Counterfeit attacks such as device id or sensor data modification

3) The information provider maliciously manipulates the access level of the data

4) Personal information leakage due to the problem of a centralized server easily duplicating a user's private information.

Above problems result in two major concerns in IoT management: Data security (data integrity) issues sent from an IoT device to another device or to a central server, and security privacy issues (device authentication) for the IoT device itself^[4]. Fig. 1 shows the conventional device access of the IoT platform. Generally, IoT devices consist of an antenna for communication, a sensor for measuring specific data, analog to digital converter, and a communication module that digitizes measured values. Communication from one IoT device include both data generated from that the device and the device's identity , and it sent to another device, a local server, or a central server. Until now, however, a number of IoT devices have not been efficiently handled security problem due to the following problems:

Fig. 1. Device Access of the Conventional IOT Platform.

1) Existing security solutions that must receive a security key from a centralized server are not used even if there is a no problem with the power budget limitation caused by the battery problem of the IoT device itself. It is necessary to create security keys on its own in various environments such as bicycles, smart homes, industries, CCTV cameras, etc.

2) Managing extremely distributed IoT device cannot be efficiently managed by a centralized server.

In this work, to solve above problem, a PUF using chaotic entropy source is proposed for IOT device management and a decentralized IoT chain using key generated by PUF is presented. Section II shows the architecture of proposed IoT chain. Section III shows the detailed implementation and circuit description of proposed PUF based on chaotic entropy source. Section IV and V show the measurement results and conclusion.

II. OVERALL ARCHITECTURE

Fig. 2 shows the overall block diagram of the proposed secure IoT chain. The public key infrastructure (PKI) is applied to resolve security issues that the existing IoT platform could not resolve.

Fig. 2. Device Access of the proposed Secure IOT Platform.

However, the key generation required for this PKI consists of power-hungry component in a typical IoT device. Therefore, we propose a low-power PUF, which has true randomness and non-replicable characteristics. PUF has 4 characteristics: Uniqueness, unpredictability, unclonability, and robustness. Therefore, the proposed IoT platform can obtain the different response $R_{i}$ ≠ $R_{j}$ (i,j ∈ devices in proposed IoT chain) for each device when the same challenge C is applied as shown in Fig. 2. More detailed implementation of PUF design will be discussed in chapter 3.

Detailed data transmission process and device authentication process is shown in Fig. 3. The proposed structure in data transmission for data integrity uses block-chain-based transmission. To resolve the security issues based on the existing public blockchain, a private key generated from PUF is used to Merkle tree in the transaction chain shown in Fig. 3(a). In the process of creating a signature from a blockchain based on the existing Merkle tree, a chain structure is combined hash function with a private key from the proposed PUF. Therefore, it can be verified that the data transmitted is valid because only authorized users with a private key can access it compared to the normal public chain.

Besides data integrity, device authentication (ID authentication) is another important security issue. For this purpose, the process which the device id is recorded in the chain is shown in Fig. 3(b).

Fig. 3. (a) Data Transmission, (b) Device Authentication Process of proposed IoT Platform.

First, a PUF based public key is generated in each device. The generated key is verified with its own private key, which creates a digital signature. A device signature that represents a device identification is generated and has been verified through the digital signature. And it registers its ID with the local edge server or block validators. This process is also recorded in every blockchain, so that unauthorized devices can be denied to access the IoT platform and prevent malicious attacks. However, there is a possibility that anonymous IoT devices can be withdrawn or participation frequently. It can be resolved by creating a group signature with a public group key that is already shared according to the application of IoT devices.

III. CHAOTIC ENTROPY SOURCE BASED PUF IMPLEMENTATION

Fig. 4 shows the principle of chaotic entropy source^[5,6]. Chaotic entropy source can be implemented in a various ways, but piecewise linear chaotic map^[5] is the most efficient implementation. First, a logistic-map is defined and random noise is applied to the initial input. At this point, the initial noise repeats amplification and analog to digital conversion, gradually producing unpredictable random streams. Fig. 5 shows the detailed implementation of chaotic map using SAR ADC. It consists of 5-b SAR ADC, residue amplifier, sample and hold switch, and post processing unit. The input VIN is first sampled into the 5-b SAR ADC input. The input is a random value consisting of the thermal noise, flicker noise, etc. of MOSFET. After passing the ADC, the input generates a residue value, and the generated residue is converted to the input value of the full range after passing through the residue amplifier. This process is repeated over again, with the ADC's conversion bit producing true random number through the postprocessing unit based on XOR. After that, a unique and non-replicable digital finger print is generated by sampling this value.

Fig. 4. Block Diagram of Chaotic Map using A/D Conversion.

Fig. 5. Overall Architecture of ADC based Chaotic Map.

Fig. 6 shows the detailed circuit implementation of Chaotic Map. The 5-b SAR ADC was implemented as a differential SAR ADC, and the capacitor DAC consists of a MOM capacitors. The gain of the residue amplifier is set to 16, and can be gain-controlled with a tail current, and an adaptive reset-comparator is used to further reduce power consumption^[6]. First, the switch $\phi_{1}$ is connected for input sampling, then the analog to digital conversion is performed. Then the $\phi_{3}$ switch is connected and the residue held in the capacitor DAC is amplified. The $\phi_{2}$ switch is then connected and the residue is stored at the front end of the residue amplifier. This process is repeated, and the converted digital bit is passed through the XOR processing unit.

Fig. 6. Detailed Circuit Implementation of Proposed Chaotic map based on SAR ADC.

IV. MEASUREMENT RESULTS

The chip microphotograph and performance summary are described in Fig. 7. The PUF was implemented by using 0.18 mm mixed CMOS process. The active chip area of the proposed chaotic entropy source is only 30 mm×150 mm. The highest data rate of the fabricated chip was 270 kbps at the supply voltage of 0.6V. The power break down of proposed entropy source is shown in Fig. 7. Capacitor DAC only consume 2 nW, digital processing unit consumes 15 nW, residue amplifier consumes 45 nW and comparator consumes 3 nW. Fig. 8 shows the measured 1M bit steam of entropy source without deterministic patterns (0=black dot, 1=white dot). NIST SP-800.22 tests^[7] are used to evaluate the randomness of output bit stream with a threshold of p-value > 0.01(significance level). The bit stream without post-processing failed due to imperfections in the implementation of the chaotic map. The imperfection in chaotic map is due to the quantization error from CDAC mismatch and comparator offset. Therefore, we performed an XOR post processing (parity-based postprocessing). Table 1 shows the NIST test result of 1M bit raw bit stream with 0.6 V supply. As shown in Table 1, when the post-processing bit width is 4, the output bit stream successfully passed all of the 15 subtests for 1M bit streams.

Fig. 7. Chip Micrograph and Performance Summary.

Fig. 8. 1M bit output stream displayed in 1024x1024 array.

V. CONCLUSIONS

In this paper, we present a PUF based on chaotic-map entropy source for secure IoT chain that not only consumes low power, but also can be integrated in IoT device with compact area. Moreover, secure IoT platform is proposed using key generated from the proposed PUF. The proposed entropy source shows world lowest 65nW power consumption while passing all NIST tests.