The new decoding chip sets a record for data transmission energy efficiency, and the power consumption is only 1~10% of similar products

On February 26, a team of scientists led by the Massachusetts Institute of Technology developed a decoder chip. Relevant research results were announced at the ongoing International Solid State Circuit Conference.

The chip uses a general-purpose decoding algorithm previously developed by MIT, which can decipher any encoded signal. Decoding data is simpler and faster than traditional technologies, and the energy consumption is only 1% to 10% of other similar hardware, or performance ratio. Other hardware is 10 to 100 times higher and can be widely used in power-critical fields such as virtual reality and 5G networks.

 

▲ Image: Christine Daniloff, Massachusetts Institute of Technology

“This is the first time anyone has broken the 1 picojoule per bit decoding barrier,” said Muriel Médard, professor of software science and engineering at NEC and a professor in the Department of Electrical Engineering and Computer Science. The energy is about the same. It’s a symbolic threshold of great significance, but it also changes the balance of the receiver, which is probably the most pressing part from an energy standpoint – we can move that from the decoder to the other part”.

In addition to Médard, the co-authors of this paper include Arslan Riaz, a graduate student at Boston University, Rabia Tugce Yazicigil, an assistant professor of electrical and computer engineering at Boston University, Ken R. Duffy, along with other academics at MIT, Boston University and Maynooth University.

Here is a simple science popularization from IT Home. Common data is transmitted in the form of bits (0 or 1), and the sender needs to encode the data and add an error correction code at the end of the data. The error correction code mentioned here is also a redundant string composed of 0 and 1, which can be regarded as the information required for hash verification. This string of information is usually stored in a specific codebook, and the decoding algorithm at the receiving end is a verification scheme designed for this special password. Obfuscated original information. Typically, each algorithm is code-specific, and most require dedicated hardware, so a device needs many chips to decode different data.

The researchers previously demonstrated a general-purpose decoding algorithm that can crack any code—GRAND (guessed random additive noise decoding), which works by guessing the noise affecting the transmission of information, then removing the noise directly from the received data, and then Check the remaining contents of the codebook. It guesses a series of noise patterns in the order in which they might appear.

In fact, the data we receive usually comes with reliability information, also known as soft information, which can help the decoder find out which parts are wrong.

It is reported that this new decoding chip is called ORBGRAND (Ordered Reliability Bits GRAND), it can use this soft information to order the data according to the probability of error in each bit. But in practice, it’s not as simple as sorting a single bit. While the least reliable section is probably the most obviously wrong, the third and fourth unreliable sections taken together are probably just as wrong as the seventh. And here ORBGRAND uses a new statistical model that can order bits in this way, because multiple bits together can be just as error-prone as a single bit.

“If your car breaks down in the middle of the road, soft information may tell you that the battery is the problem. But if it’s not just the battery, such as the battery and the engine at the same time, it’s troublesome,” Médard said, which is a How does a rational thinking person do it – although it may be that these two parts fail together, you will start troubleshooting from the most likely place to go wrong, and then look down, and finally find something that is unlikely to happen.

This, they argue, is a more efficient approach than conventional decoders. It is reported that traditional decoders only pay attention to the code structure, and it is usually designed for the worst case. “With a traditional decoder, you take out the blueprints of the car and check every single part, and eventually you find the problem, but it takes you a long time,” Médard explained.

According to reports, as long as a key code is found, ORBGRAND will stop sorting, which is usually very fast. In addition, the chip uses parallelized logic to generate and test multiple noise patterns simultaneously to find such critical codes more quickly.

https://oaicon.com/index.php/2023/03/07/the-new-decoding-chip-sets-a-record-for-data-transmission-energy-efficiency-and-the-power-consumption-is-only-110-of-similar-products/

When they compared this approach to other chips, ORBGRAND consumed only 0.76 picojoules per bit when decoding with maximum accuracy, smashing previous performance/energy records. This also means that the energy consumption of ORBGRAND is only one-tenth or even one-hundredth of that of other equipment.

One of the biggest challenges in developing chips is reducing energy consumption, Médard said. And ORBGRAND is now so energy efficient that other processes that researchers hadn’t focused on before, such as checking codewords in codebooks, consume most of the energy.

“Right now, this inspection process, like opening a car to see if it works, is the hardest part. So we need to find more efficient ways to do that,” the team is also exploring ways to change the modulation of the transmission , so that they can take advantage of the increased efficiency of the ORBGRAND chip.