Sukumar Nandi

This is an erratum for the article "AxLEA: Approximate ARX-based Lightweight Encryption Algorithm for Resource Constrained Devices" published in ACM Transactions on Embedded Computing Systems 25, 4, Article 61 (July 2026), 23 pages.

In multilingual higher education contexts, English often functions as the primary language of instruction, academic communication and instructional interaction, even for students who acquired English as a second language. This raises an important question for applied bilingualism research, that if, sustained English medium academic exposure be associated with faster task-specific processing in English than in the first language? The present study examines this question in Bengali-English and Hindi-English bilinguals enrolled in the same English medium academic environment. Participants completed a picture-word matching task in which visual stimuli were paired with written labels in their respective L1 and in English. Response time and accuracy were used as behavioural indices of task-specific processing efficiency. The results showed that participants in both bilingual groups responded faster in English than in their respective L1, with English also showing a small accuracy advantage. The English response time advantage was larger in the Hindi-English group than in the Bengali-English group, while between group accuracy differences were not statistically robust. These findings suggest that English medium academic exposure may be associated with task specific processing advantage for English in speeded lexical-semantic verification. However, the results are not interpreted as evidence of fixed or global English dominance. Instead, they suggest that language strength is contingent on the context of institutional exposure, task demands and processing. The study contributes to research on bilingualism in multilingual educational settings by showing how behavioural task performance can reveal processing asymmetries that are not reducible to acquisition order alone.

Pseudo Random Functions (PRFs) and lightweight message authentication are key components of embedded and Internet of Things (IoT) security architectures. These systems operate under strict constraints on latency, energy consumption, and hardware area, which require hardware-efficient implementations. SipHash is a widely used ARX-based PRF and MAC optimized for short inputs. However, its hardware implementations are frequently constrained by the latency and power consumption of 64-bit modular additions in the round function. This paper presents HAxSH, a hardware-oriented and performance-focused derivative of the SipHash that uses approximate arithmetic as an architectural optimization. HAxSH retains the round function structure of SipHash while selectively replacing exact 64-bit additions with carefully designed approximate adders to reduce the critical path and switching activity. Two approximate adder architectures are designed by approximating carry propagation and sum computation, and are integrated into the ARX round function. The impact of approximation is empirically evaluated using NIST STS, ENT, DieHarder, and avalanche testing under a classical adversary model. In addition, MILP-based automated linear and differential cryptanalysis is performed to bound characteristic propagation across rounds. Hardware performance is evaluated through ASIC synthesis in 90nm technology. Compared to SipHash and Chaskey, HAxSH achieves up to 6× lower latency, approximately 10% area reduction, 60% lower power consumption, and a throughput of 1.95 Gbps for 128-bit messages. These results suggest that approximate arithmetic can be a viable hardware acceleration technique for ARX-based PRFs targeting lightweight message authentication in highly constrained environments.