Table 1 Comparison with existing on-chip ONN works

Publication	Pixels	Computing speed	Computing density	In-sensor	Incoherent light	MMI	Application
X., X. et al.²⁷ Nature, 2021	500 × 500	1.785 TOPS	–	×	×	×	handwritten digit recognition (HDR)/image processing
F., J. et al.²⁶ Nature, 2021	128 × 128	4 TOPS	1.2 TOPS/mm2	×	×	×	HDR/edge detection
A., F. et al.¹¹ Nature, 2022	5 × 6	0.27 TOPS	3.5 TOPS/mm2	×	×	×	low-resolution image classification
F., T. et al.²⁴ Nat. Commun., 2023	28 × 28	13.8 POPS	–	×	×	×	HDR
M., X. et al.³¹ Nat. Commun., 2023	28 × 28	0.27 TOPS	25.48 TOPS/mm2	×	×	×	HDR
B., B. et al.³² Nat. Commun., 2023	250 × 250	–	1.04 TOPS/mm2	×	×	×	HDR/edge detection
D., B. et al.³³ Nature, 2024	28 × 28	0.108 TOPS	–	×	×	×	HDR
Ours	400 × 533	Adaptive^a	Adaptive	√	√	√	complex tasks in the real world: face anti-spoofing and disease diagnosis

^aOCL performs analog computing. It has adaptive computing speed that can always meet the imaging speed.

Quick links

Search