Security and Communication Networks

Research Article

Improve Neural Distinguishers of SIMON and SPECK

Table 1

Comparison of neural differential distinguishers.


Block cipher	Source of neural distinguisher/input difference	Round	Accuracy (%)

SIMON32/64	Reference [10]¹	8	82.2
	Reference [17]²	9	59.07
	Reference [13]³	9	63.73
	Section 3	9	59.77
	Section 4	9	82.27
	Section 4	10	61.09

SIMON48/96	Reference [17]²	9	50.22
	Reference [10]¹	10	53.49
	Section 3	10	57.89
	Section 4	10	81.40
	Section 4	11	61.43

SIMON64/128	Reference [17]²	10	58.61
	Section 3	11	59.72
	Section 4	11	73.79
	Section 4	12	69.57

SPECK32/64	Reference [7]	7	61.6
	Reference [12]³	7	70.74
	Section 4	7	88.19
	Reference [7]	8	51.40
	Section 4	8	56.49

SPECK48/96	Reference [15]	5	—⁴
SPECK48/96	Section 4	7	63.43

SPECK64/128	Section 4	8	63.20

¹We train neural distinguishers using Benamira et al. ’s method presented in [10]. ²In [17], Abed et al. constructed differential characteristics of SIMON. We train neural distinguishers by choosing the input differences in [17]. ³We choose the highest-accuracy neural distinguisher in [12, 13]. ⁴ Chen et al. used 5-round neural distinguisher to attack SPECK48/x, but the accuracy was not presented in [15].