Error correction via artificial intelligence

1 . An apparatus, comprising:

a memory device;

an augmented reality (AR) display coupled to the memory device; and

artificial intelligence (AI) circuitry coupled to the memory device and configured to:

execute an AR AI workload associated with the AR display;

receive an error correction model that emulates execution of error correction code (ECC),

wherein the error correction model is trained by a host system to correct random errors introduced to execution of the AR AI workload in hazardous conditions via inputting a plurality of weights of the AR AI workload with random errors; and

execute the trained error correction model as a layer of the AR AI workload, to perform software-based error correction.

2 . The apparatus of claim 1 , wherein the error correction model is trained such that inputting the plurality of weights with random errors to the AR AI workload and executing the error correction model causes the AR AI workload to yield a same output as the AR AI workload would without the random errors,

wherein the plurality of weights comprise weights of an untrained error correction model.

3 . The apparatus of claim 1 , wherein the error correction via execution of the error correction model as a layer of the AR AI workload is carried out subsequent to execution of ECC by the memory device.

4 . The apparatus of claim 1 , wherein the memory device comprises dynamic random access memory (DRAM); and

wherein the AI circuitry comprises a deep learning accelerator (DLA).

5 . The apparatus of claim 1 , wherein the memory device comprises a memory array on-chip with the AI circuitry such that the memory device comprises a system-on-chip (SoC).

6 . The apparatus of claim 1 , wherein the AR display comprises a component of personal protection equipment (PPE); and

wherein the hazardous conditions comprise a radioactive environment.

7 . A method, comprising training, by a host system, an error correction model to emulate execution of error correction code (ECC) to correct random errors from operating an augmented reality (AR) device in hazardous conditions via input of weights of an AR artificial intelligence (AI) workload with errors;

subsequent to training the error correction model, communicating the error correction model from the host system to the AR device;

executing the AR AI workload on a deep neural network (DNN) implemented on AI circuitry of the AR device; and

performing software-based error correction associated with executing the AR AI workload via executing the error correction model on the DNN.

8 . The method of claim 7 , wherein executing the AR AI workload comprises executing a vision-based AI model to perform an AR function; and

wherein performing the software-based error correction comprises executing the error correction model to emulate execution of the ECC to correct random errors introduced to execution of the vision-based AI model by the hazardous conditions.

9 . The method of claim 7 , further comprising, responsive to operating the AR device in non-hazardous conditions, performing hardware-based error correction associated with executing the AR AI workload via error correction circuitry of the AR device.

10 . The method of claim 9 , further comprising, responsive to operating the AR device in the hazardous conditions subsequent to operating the AR device in the non-hazardous conditions, switching from performing the hardware-based error correction to performing the software-based error correction.

11 . The method of claim 10 , further comprising, responsive to operating the AR device in the non-hazardous conditions subsequent to operating the AR device in the hazardous conditions, switching from performing the software-based error correction to performing the hardware-based error correction.