Practical solutions in fully homomorphic encryption: a survey analyzing existing acceleration methods

Cybersecurity

Table 7 Comparison of acceleration schemes based on FPGA

References	Year	Availability	Data storage optimization	Homomorphic parameters tuning	Accelerated object	Acceleration result
Sujoy Sinha Roy (2019)	2019	+ + +	×	×	BFV	13 ×
Mert and Öztürk (2020)	2019	+ + +	×	×	BFV	7–12 ×
Riazi et al. (2020)	2020	+ +	×	√	CKKS	164–268 ×
Kim et al. (2020)	2020	+ + + +	√	×	NTT	118 ×
Rashmi (2020)	2020	+ + +	×	×	Polynomial Multiplication	2950–4200 ×
Turan (2020)	2020	+ + +	√	×	BFV	5 ×
Serhan et al. (2021)	2021	+ +	×	×	Bootstrapping	*
Fadhli et al. (2021)	2021	+ +	×	×	BFV	3.85 ×
Ye et al. (2021)	2021	+ + +	√	×	BFV	3.4–6.7 ×
Xin and Zhao (2021)	2021	+ + + +	×	×	CKKS	26.04 ×
Syafalni et al. (2022)	2022	+ +	×	×	Polynomial Multiplication	38.5 ×
Yang et al. (2022a)	2022	+ +	√	×	BFV	5.6 ×
Han et al. (2022)	2022	+ + +	×	×	Key-switching	1.6 ×
Agrawal et al. (2022)	2022	+ + +	√	×	Bootstrapping	533 ×
Ye and Kannan (2022)	2022	+ +	√	×	Bootstrapping	16.5 ×
Yang et al. (2022b)	2022	+ +	√	×	NTT	#

*(Serhan et al. 2021) uses less hardware resource to achieve linear performance scaling with up to 16 vector lanes about matrix–vector operations in bootstrapping of TFHE. And it doesn’t achieve significant acceleration
#(Yang et al. 2022b) also uses less hardware resource to implement FHE and doesn’t give the acceleration result in terms of latency