AES Implementation and Performance Evaluation on 8-bit Microcontrollers

The sensor network is a network technique for the implementation of Ubiquitous computing environment. It is wireless network environment that consists of the many sensors of lightweight and low power. Though sensor network provides various capabilities, it is unable to ensure the secure authentication between nodes. Eventually it causes the losing reliability of the entire network and many secure problems. Therefore, encryption algorithm for the implementation of reliable sensor network environments is required to the applicable sensor network. In this paper, we proposed the solution of reliable sensor network to analyze the communication efficiency through measuring performance of AES encryption algorithm by plaintext size, and cost of operation per hop according to the network scale.

💡 Research Summary

The paper addresses the pressing need for secure communication in resource‑constrained sensor networks by implementing the Advanced Encryption Standard (AES) on an 8‑bit microcontroller and evaluating its performance under realistic conditions. The authors begin by outlining the characteristics of typical wireless sensor networks: large numbers of lightweight nodes, limited battery capacity, modest flash and RAM sizes, and the inability of many existing deployments to provide robust authentication and confidentiality. While prior work has demonstrated AES on 16‑bit or 32‑bit platforms, there is a notable gap in empirical data for ultra‑low‑power 8‑bit devices, which are still widely used in cost‑sensitive deployments.

The target hardware is an Atmel AVR ATmega128 running at 8 MHz, equipped with 128 KB of flash and 4 KB of SRAM. AES‑128 is implemented in Electronic Codebook (ECB) mode using a hybrid of C and hand‑optimized assembly. The key schedule is pre‑computed during initialization to avoid repeated overhead, and the substitution boxes (S‑Box and inverse S‑Box) are stored as constant arrays in flash, limiting RAM usage to under 200 bytes. The authors deliberately choose ECB for its simplicity in benchmarking, while acknowledging its known pattern‑leakage weakness.

Performance is measured along two axes. First, encryption latency is recorded for plaintext sizes ranging from one block (16 bytes) to sixteen blocks (256 bytes). Each size is encrypted 100 times, and the average cycle count and energy consumption are reported. The results show an almost perfectly linear relationship between block count and cycles: 1 block ≈ 1,560 cycles, 8 blocks ≈ 12,400 cycles, and 16 blocks ≈ 24,800 cycles. At 8 MHz, this translates to 0.195 ms per block, with an energy cost of roughly 3.9 mJ for a 128‑byte payload—well within the budget of typical sensor node batteries.

Second, the authors model per‑hop computational cost in a multi‑hop routing scenario. Assuming that each hop must both encrypt outgoing data and decrypt incoming data, the total end‑to‑end delay scales linearly with hop count: a single hop adds about 0.78 ms, while a ten‑hop path adds roughly 7.8 ms. The cumulative energy consumption for a ten‑hop transmission of a 128‑byte packet remains under 40 mJ, indicating that even with multiple relays the overhead is modest compared to the overall power budget of a sensor node.

The discussion highlights several practical considerations. ECB’s deterministic mapping can reveal data patterns, so the authors recommend pairing the implementation with more secure modes such as CBC, CTR, or GCM for production systems. Key management is another challenge; the limited RAM makes dynamic key exchange difficult, suggesting the use of pre‑shared keys or lightweight authenticated key exchange protocols (e.g., DTLS with elliptic‑curve cryptography). The paper also notes that while software‑only AES is feasible, hardware acceleration blocks present in newer microcontrollers could further reduce latency and energy consumption.

In conclusion, the study demonstrates that AES‑128 can be efficiently executed on an 8‑bit microcontroller with acceptable latency and energy costs, making it a viable security primitive for low‑cost sensor networks. The authors propose future work that includes benchmarking against hardware‑accelerated AES, evaluating alternative cipher modes, and conducting field trials to validate the energy model in real deployment scenarios.