. 2024 Jan 2;15:129. doi: 10.1038/s41467-023-44620-1

Table 2.

The comparison results with traditional microelectronic chips, which function as hardware accelerators for computation

	This work	Ref. ⁵⁸.	Ref. ⁵⁹.	Ref. ⁶⁰.	Ref. ⁶¹.	Ref. ⁶².	Ref. ⁶³.
Hardware	Memristive passive crossbar array	GPU (Titan X)	FPGA (Virtex 7)	FPGA (Arria-10)	FPGA (Zynq-7000)	ASICs (YodaNN)	ASICs (Google TPU)
Energy efficiency	4.35 TOPS/W	98 GOPS/W	935 GOPS/W	30 GOPS/W	14.3 GOPS/W	8.6 TOPS/W	2.3 TOPS/W

This work

Ref. ⁵⁸.

Ref. ⁵⁹.

Ref. ⁶⁰.

Ref. ⁶¹.

Ref. ⁶².

Ref. ⁶³.

Hardware

Memristive passive crossbar array

GPU (Titan X)

FPGA (Virtex 7)

FPGA (Arria-10)

FPGA (Zynq-7000)

ASICs (YodaNN)

ASICs (Google TPU)

Energy efficiency

4.35

TOPS/W

GOPS/W

935

GOPS/W

14.3

GOPS/W

8.6

TOPS/W

2.3

TOPS/W