A novel photonic device that employs light for computation has been created by researchers. Although I’ve discussed a lot of photonic chips, this optical device is unique, and today I’ll describe how it functions and how they were able to accomplish this performance multiplied by a thousand. Let me clarify: when I think of light, the first image that comes to me is of warm, golden sunlight. Color is a wavelength of an electromagnetic wave, some of which are visible to us.
The majority of them are beyond our reach, yet photons—tiny energy packets—are what makeup light. As we will see later in the video, the main difference between photonic and electronic computer processors is that photonic use light instead of electrons for computation, and we can compute data on the fly without having to stop the data as we do in classical computing, which processes the data as it travels. This computation on the fly is in the range It’s really quick and has nearly no latency—you know, we got from microseconds to none when we moved from vacuum tubes to transistors—but the Fatoni computer gluing enables us to cut it down to F seconds.
The first revolution in photonic computing occurred with the invention of the laser, which is a device that transforms random unfocused photons into a powerful focused beam. Charles Towns, who is regarded as the father of avionics, received a Nobel Prize for this invention more than 60 years ago, but at the time there were few practical uses for the laser, whereas today we use it almost every day multiple times, from printers and scanners to quantum computers and extreme ultraviolet lithography machines that I used to fabricate every silicon chip that exists today, such as AMD CPUs or Nvidia GPUs.
simply not be possible without these devices, and without the laser—just consider the Laser Quest—which is undoubtedly worthy of a Nobel Prize. In addition, the creation of optical fiber for communication was another significant milestone in photonics, and it was led by Charles Kunga, who was also later awarded the Nobel Prize in Physics for his contributions. Kunga created a unique quir glass material that was sufficiently transparent to prevent photons from being sent over enormous distances—hundreds of hundreds of kilometers—so that information fiber, and nothing in the world moves more quickly than that.
Taichi Chip
The next significant milestone in photonics was the development of silicon photonics because, at this point, we’ve already achieved and learned a lot with silicon, including how to manufacture silicon chips on a massive scale. Then, there was this idea to take all we know about semiconductor design and Manufacturing and integrate silicon with photonic and use photons. They then advanced it and succeeded in sending multiple signals at the same time through this same fiber. This was made possible due to the beautiful property of light that it has different colors or different wavelengths of light that don’t interfere with each other. This of course also applies to computing.
not only for data transmission but also for computing itself, as this allows us to utilize the speed of light and massive parallelism—what a fantastic idea, right? Additionally, a very promising field of forensic quantum computing has started to emerge, enabling the dream of quantum computing—quantum computing at room temperature. If you would want me to do a follow-up episode on it, please let me know in the comments below. Don’t forget to subscribe to the channel so you don’t miss it. Now, let’s see how this TaiChi chip functions. It’s a photonic chipset, which means that it combines many chips with various functionality into a single unit.
a single package if all of the photonic chips are taken into account. are the two primary methods for creating them. It typically employs either light interference or diffraction to compute, but this device goes one step further and mixes the two in a rather intriguing way. Let’s start with light interference from automobiles. Given how well my discussion on light interference went in the previous video, I should keep riding this wave. Jokes aside, I hope you’ve pardoned me anyway, as a lot of people figured out what I was talking about since, as you know, light is a wave and waves interact.
Photonic Logic Gates
into disagreement or overlap When I take two sinusoids and shift their pH 180°, we have a phenomenon called destructive interference. In this case, the two waves cancel each other out, resulting in a decrease in light intensity or even complete darkness. At the most basic level, the resting wave of two sinusoids in the same phase will have twice the amplitude and we will be able to notice it more because it will just be brighter. These days, a lot of photonic devices use the idea of light interference to carry out linear operations. So, how does it operate, or more simply put, how can we carry out basic logic operations?
Logic gates, which are tiny switches made of transistors, are typically used to implement light logic operations on chips. An electronic circuit is made up of billions of switches that alternate between on and off states, and each gate has a logical operation behind it. For instance, when we use an end logic gate, the output that results depends on the combination of the input values.
When the inputs are one and zero, zero and one, or Z and zero, the output is zero. In electronics, all zeros and ones are represented by voltages by voltage. Since both inputs are one, we obtain one at the output. level; that is, if the voltage is above a given value, a threshold, it is one below zero. In photonics, this process is similar, but instead of voltages, we measure the light intensity at the output; if it is above a given value, it is one; if not, it is zero. This object is a
logic end gate It features one output and two inputs. The first thing we do is utilize interference in this optical combiner to ensure that the strong field only occurs when both inputs are turned on. For the subsequent Gates that listen to this Gates to comprehend the signal, the second stage is this microing, which allows us to create a strong distinction between the on and the off level. I suppose you can imagine how much research this requires before we talk about how we can use diffraction for computing this light.
Computing with Diffraction
This can be seen as the bending of light as it passes through the opening. This method is easier to understand if you think about what waves are diffracted photonic diffraction, which is essentially light waves bending around the obstacles and You guessed it—it may also be used for logical operations.
Let’s imagine that the amplitude of the light is controlled using code. After that, light moves from the input layer to the output layer via the diffraction layer, with each defraction layer standing in for a distinct meta surface linked to a particular logic process. An example of a sophisticated logic gate in this scenario has an input of one and zero and an output of one. Yes, but it works effectively; yet, the main drawbacks of this approach.
This study is fascinating because both processes are hardcoded into the surface and cannot be altered. Instead, they combined the best elements of both strategies—light diffraction and light interference—into a single TaiChi chip. Typically, photonic computing’s input data coding step is a weak point because the conversion from digital to analog—that is, from digital to light—is a rough process that is difficult to make as accurate as digital signal processing. However, in this case,
Since it’s a known procedure, they may use diffraction to accomplish it by using diffraction surfaces. In this instance, information was encoded into optical patterns using light diffraction on a semiconductor. This is the best that can be achieved. For the computing parts, light interference is used. You see, when we discuss any computation related to artificial intelligence, we primarily mean operations involving matrices. All that happens is that light passes through the diffraction layers and is encoded on the fly at practically no energy and in a massively parallel way.
How Taichi Chip Works
While the activation function is another nonlinear component that cannot be disregarded, the majority of operations in neural networks involve matrices. All information is represented by matrices. For instance, to do image recognition and determine what is on it, we must multiply many matrices and add the results for this chip to carry out the multiply-accumulation operation using photonics. The input information is first converted into the amplitude of light by them, and this light is subsequently sent via a so-called mock Xander interferometer:
This gadget separates light into two pathways, which it then rejoins and splits again, as the name would imply. We multiply the input matrix through these cells with the weight, add it up with the optical attenuator, and then decode the result through the diffraction layer. The most intriguing thing about this is that you can change the path that the light takes through one of these paths to multiply the light by a specific value.
Results
Since classical computers operate serially, one operation follows another. In this scenario, to perform multiple operations in parallel, we would need multiple GPUs. However, photonic computing can perform multiple operations in parallel by nature because it can compute at multiple wavelengths of light simultaneously, which is brilliant. What sets this chip apart from previous discussions of photonic chips is that it can perform more operations than
sophisticated computing assignments If we use the AI application as an example, the new Taii chip can manage up to 14 million parameters, whereas prior optical processors could only handle hundreds of thousands. parameters, which implies it can carry out considerably more complex tasks.
For instance, when it comes to image classification, the chip was tested using a data set that included thousands of handwritten characters from 50 different alphabets, and it completed the task with 92% accuracy, which is fantastic Because photonic chips are analog, and one of the primary issues with analog is accuracy, as I recently read in a study report regarding light-matter chips.
And from what I recall, the accuracy was around 80% then. However, as you can see from their study, they have already attained 92%. Researchers have put this TA chip to use on generative AI tasks, and the findings are fascinating since they demonstrate how well the photonic technology truly functions. The chip can create music in the manner of German composer
Bak and it can already create landscapes in the vein of Vincent van Gogh and Edward Mo, the latter of whom is renowned for his well-known painting This Cream. They measured the chip’s performance during these operations, and it can achieve 163 actions per second per watt. According to the paper’s performance comparison, the chip is reportedly more energy-efficient than any of the most recent Nvidia h100 gpus.
According to an earlier study, it was around 100 times more energy-efficient and ten times more area-efficient than earlier photonic chips. As you may know, the area is one of the main issues with photonics. Although the chip itself is fairly compact, the entire system is not because they used a laser source, which may be as large as a table. I’m not sure which specific laser they used, but generally speaking, it can consume up to one kilowatt, negating all the efficiency gains we’ve just discussed.
are generally rather large. For instance, the cell we discussed has dimensions of about 50 micrometers, making it enormous in comparison to the world of semiconductor chips, which have transistors in the nanometer range and relatively inexpensive logic that can fit in the area of a single photonic device. Of course, many faps and researchers are working on further photonic components, but it’s still far from the conventional computer chips, as we previously saw.
It’s a terrific alternative method of scaling computing without just growing the number of components. Performing many calculations at different wavelengths of light can make up for it. work first, they created a chipset that can run a neural network with millions of parameters. Based on the paper, I can see that they have plans to scale this chip further. Yes, it’s obvious that this chip is only being used for inference applications at this time and won’t be on your home computer tomorrow, but it’s still a promising development.
Source:Anastasi In Tech
