The Electronics and Telecommunications Research Institute (ETRI) of Korea announced on Sept. 6 that it has developed a dedicated computational accelerator chip capable of high-speed processing of homomorphic encryption, which is drawing much attention as a next-generation cryptographic technology.
Unlike conventional password encryption, homomorphic encryption is a technology that calculates encrypted data directly without re-solving it. There is no need to use a secret key to decrypt encrypted data, so there is no risk of exposing the secret key or the original data. It is also quantum resistant which means that it cannot be cracked by quantum computing.
Once this technology is commercialized, it will be used in data center servers for cloud services and artificial intelligence semiconductors that can guarantee privacy, ETRI researchers believe. Research on chips and software development platforms for homomorphic encryption accelerators is also expected to gain momentum.
This technology is expected to enable data to be encrypted and directly applied to various convergence services where security and statistical and artificial intelligence applications are required simultaneously so it can apply to the defense, public, medical, financial, and industrial sectors, among others. It can befit cloud computing and database management system (DBMS) companies, fabless companies, server-mounted homomorphic encryption accelerator developers, data privacy service companies, and companies that utilize sensitive data such as public and defense data.
This fully homomorphic hardware computational accelerator chip developed by ETRI is a dedicated hardware computational processing unit suited to the unique characteristics of homomorphic encryption where a ciphertext is represented by polynomials of tens of thousands of degrees or more degrees with coefficients of thousands of bits.
They perform fast operations such as addition, subtraction, multiplication, and division between high-degree polynomials while data is still encrypted. The principle is that the encrypted data is directly computed and only the result can be decrypted to expose it data. No one can see the original data, so data privacy is guaranteed.
When data is turned into ciphertext, it becomes a polynomial tens of thousands of times bigger, which increases the amount of computation exponentially. Therefore, it was quite a big challenge to utilize homomorphic encryption because it is not suitable for computing with a central processing unit (CPU) or a graphics processing unit (GPU) that processes 64-bit data in terms of structures, as they take too much time.
This research project solved these problems and provides a core technology for a homomorphic processing unit (HPU) that can utilize homomorphic encryption, the researchers explained.
Based on this technology, the researchers plan to implement and commercialize a system on chip (SoC) semiconductors dedicated to homomorphic encryption in the future. Furthermore, they will develop a platform that extends a fully homomorphic hardware accelerator that can infer a video with homomorphic encryption to an artificial intelligence model with homomorphic encryption in real time with an AI model powered by a homomorphic encryption SoC.