The mathematics behind the FHE, developed in 2009, involves operations that require resource- and time-consuming calculations. The solution lies in hardware that is natively capable of processing these very large numbers. As a computer architect, Reage designs chips that can do very special things to run much faster than general-purpose hardware. Now you have been given the task of laying the foundations for this FHE chip.
“To enable homomorphic encryption, my task is to grasp the building blocks of this unit – not just the memory, but many parallel functions, as well as LAWS units designed by Mihalis – and customize, organize, and distribute them on a chip to a specific area to maximize performance, which is ultimately the basic barrier to this encryption process. there would be no question, everyone would use FHE for the simple reason that we would never have to decrypt the data. My job is to bring about this performance. “
Function-specific hardware (think such as a graphics card or processing units designed for machine learning operations) is not new. In the case of FHE, the key lies in the bits that the computer uses to represent the numbers for the quick calculations it performs. “The architecture of a typical processor for 32-bit operations would take too long to process the huge numbers needed for FHE and use too many resources to do so. It could be done, but it would be incredibly slow. It would be six orders of magnitude slower than uncoded computing. Our job is to reduce the deceleration to a maximum of one order of magnitude, “explained Maniatakos.
He explained that at 32 bits, the largest number that a chip can process is 32 times the power of four billion, which is enough for most applications. The problem is that with FHE, the numbers processed are much larger than 32 bits, are in the order of thousands of bits, and cannot be represented natively, so the computer has to divide these numbers, which involves a lot of unnecessary work (overhead).
“Imagine that there is a list of numbers, say one to ten. Suppose we first encrypt this data so that this very small list of numbers is transformed into two lists where each item is 10-100 times larger. And all of a sudden, right at the beginning, we increase the size of the numbers by 20 to 200. Now, if you want to perform multiplication functions on this bloated data, you will not only do multiplication, but you will also handle many other functions that transform the data representations. running metafunctions actually takes longer than running actual multiplication, “Reagen explained.
” The six The main challenge in designing thin architectures is how to balance and process all these processes, and working with the person designing the functional hardware units in LAWS is critical to making the performance a reality, ”the researcher pointed out. Other universities and research institutes (Carnegie Mellon University, SpiralGen, Drexel University, TwoSix Labs) will join the development
Hardware, software, tests, curiosities and colorful news from the IT world by clicking here!
