Computing system and computing method with de-identification architecture and de-identification circuit

Abstract

This disclosure proposes a computing system with a de-identification architecture. The computing system with a de-identification architecture includes a de-identification circuit and a decision circuit. The de-identification circuit includes an analog memory-based computing circuit, a first encryption circuit, and a second encryption circuit. The analog memory-based computing circuit is configured to generate computational data based on input data. The first encryption circuit is configured to perform an input-sensory shift operation on the computational data based on the input data to generate input-sensory encrypted data. The second encryption circuit is configured to perform an analog encryption amplification operation on the input-sensory encrypted data based on the analog physically uncopyable function of the second encryption circuit to generate de-identified data. The decision circuit includes multiple feature enhancement models, an analog enhancement circuit, and a voting circuit. The multiple feature enhancement models are configured to receive the de-identified data and, based on each of the multiple feature enhancement models, perform feature enhancement operations on the de-identified data to generate multiple feature-enhanced data. The analog enhancement circuit is configured to perform analog enhancement amplification operations on multiple feature enhancement data at multiple time periods based on the analog physical non-replicable function of the analog enhancement circuit, thereby generating multiple candidate data. The voting circuit is configured to determine the decision result based on the multiple candidate data.

Claims

1. A computing system with a deidentification architecture, comprising: a deidentification circuit, including: an analog memory-based computing circuit configured to: generate computing data based on input data; a first encryption circuit configured to: perform an input-sensory shift operation on the computing data based on the input data to generate input-sensory encrypted data; and a second encryption circuit configured to: perform an analog encryption amplification operation on the input-sensory encrypted data based on an analog-physical non-copyable function of the second encryption circuit to generate deidentified data; and a decision circuit, including: a plurality of feature enhancement models configured to: receive the deidentified data; and perform a feature enhancement operation on the deidentified data based on each of the plurality of feature enhancement models to generate a plurality of feature-enhanced data; an analog enhancement circuit configured to: perform an analog enhancement amplification operation on the plurality of feature-enhanced data at multiple time periods based on an analog-physical non-copyable function of the analog enhancement circuit to generate a plurality of candidate data; and a voting circuit configured to: A determination result is made based on the multiple candidate data.

2. The computing system of claim 1, wherein the input data comprises multiple sensing data from multiple sensors, the analog memory in-cell computing circuitry comprises multiple weight blocks, and the analog memory in-cell computing circuitry is configured to: input the multiple sensing data into the multiple weight blocks respectively.

3. The computing system as claimed in claim 1, wherein the input sensing displacement operation is configured to: change a displacement applied to the computing data based on a change in the input data.

4. The computing system as described in claim 1, wherein the degree of data association of the input-aware encrypted data is less than the degree of data association of the computing data.

5. The computing system as claimed in claim 1, wherein the analog encryption amplification operation is configured to: amplify the analog physical uncopyability of the second encryption circuit and apply the amplified analog physical uncopyability to the input sensing encryption data.

6. The computing system as described in claim 1, wherein the degree of data difference of the deidentified data is greater than the degree of data difference of the input-aware encrypted data.

7. The computing system as claimed in claim 1, wherein multiple feature enhancement models are pre-trained separately for the multiple sensors that serve as the data source of the input data.

8. The computing system as claimed in claim 7, wherein the feature enhancement operation is configured to amplify the data features obtained by the plurality of sensors.

9. The computing system of claim 1, wherein the analog enhancement amplification operation is configured to: amplify the analog physical non-reproducible function of the analog enhancement circuit at different ratios during the plurality of time periods, and apply the amplified analog physical non-reproducible function to the feature enhancement data to generate the plurality of candidate data.

10. The computing system as claimed in claim 1, wherein the voting circuit is configured to determine the determination result based on the common features of the plurality of candidate data.

11. A deidentification circuit, comprising: an analog memory in-cell arithmetic circuit configured to: generate a computational data based on input data; a first encryption circuit configured to: perform an input-sensory shift operation on the computational data based on the input data to generate an input-sensory encrypted data; and a second encryption circuit configured to: perform an analog encryption amplification operation on the input-sensory encrypted data based on an analog physically uncopyable function of the second encryption circuit to generate deidentified data.

12. The deidentification circuit as claimed in claim 11, wherein the input sensing displacement operation is configured to: change a displacement applied to the computational data based on a change in the input data.

13. The deidentification circuit as described in claim 11, wherein the data correlation degree of the input sensing encrypted data is less than the data correlation degree of the computational data.

14. The deidentification circuit as claimed in claim 11, wherein the analog encryption amplification operation is configured to: amplify the analog physical uncopyability of the second encryption circuit and apply the amplified analog physical uncopyability to the input sensing encrypted data.

15. The deidentification circuit as claimed in claim 11, wherein the degree of data difference of the deidentified data is greater than the degree of data difference of the input-aware encrypted data.

16. A computational method with a deidentification architecture, comprising: generating computational data based on input data via an analog memory computational circuit; performing an input-sensory shift operation on the computational data based on the input data via a first encryption circuit to generate input-sensory encrypted data; performing an analog encryption amplification operation on the input-sensory encrypted data via a second encryption circuit based on an analog physical non-copying function of the second encryption circuit to generate deidentified data; performing a feature enhancement operation on the deidentified data based on each of a plurality of feature enhancement models to generate a plurality of feature-enhanced data; performing an analog enhancement amplification operation on the plurality of feature-enhanced data at a plurality of time periods via an analog enhancement circuit based on an analog physical non-copying function of the analog enhancement circuit to generate a plurality of candidate data; and determining a determination result based on the plurality of candidate data via a voting circuit.

17. The computational method as described in claim 16, wherein the input-sensing displacement operation is configured to: change a displacement applied to the computational data based on a change in the input data.

18. The operation method as described in claim 16, wherein the degree of data association of the input-aware encrypted data is less than the degree of data association of the operation data.

19. The operational method as described in claim 16, wherein the analog encryption amplification operation is configured to: amplify the analog physical uncopyability of the second encryption circuit and apply the amplified analog physical uncopyability to the input sensing encryption data.

20. The operation method as described in claim 16, wherein the degree of data difference of the deidentified data is greater than the degree of data difference of the input-aware encrypted data.

Images