five years ago, Eric Aguler Was fed up.
He worked on Lidar and other sensors for years in Tesla and Google X, but technology always seemed very expensive and more importantly, incredible. They replaced the lidar sensor when they broke – which were all often, and randomly and developed complex calibration methods and maintenance routines simply to develop them and operate cars.
Therefore, when he arrived at the end of his rope, he invented a more strong technique – which he calls “the most powerful micromachin” ever.
Aguilar and his team in startup Omnipotent sensor The developed new micro-electro-mechanical system (MEMS) technology that claims that it can produce more force per unit area than any other. By supplying new electrical levels to the micrometer, the technology is capable of steering properly to the laser beams of the lidar, even hazing elements and open road explosions and bangs. With chips under testing by auto-industry customers, Omnitron is now modifying technology to reduce power consumed by AI data centers.
Lidar, a scanning and detection system that uses lasers to determine how far the objects are, are often adopted by self-driving cars to find and navigate obstacles. Even market expects for Lidar The annual increases 13.6 percent annuallyThe use of lidar in the automotive industry has been relatively stable in recent years, says the agoler, in the part because the lifetime of technology is so low.
Bumpy roads and severe environmental conditions vibrations are the biggest reliability killers for automotive lidar, say Moe LeeThe WHO studies the photonic system at the University of Washington. Optical alignment within the lidar package is delicate over self-driving cars-a poor paving job directors can physically replace where the mirrors sit in the house, possibly the beam is incorrectly added wrongly and fails the system. Or temperature ups and downs can cause parts to expand or contract with the same unfortunate result, he explains.
The leader wondered which part was most often broken and the culprit was found to be scanner, the part responsible for engulfing small mirrors that direct the laser beam out of the residence of the system. He wanted to make scanners that could face the tough conditions of the lidar face, and silicon flexor stood out as a solution. These structures act as springs and allow for careful control of mirrors within the lidar system, as standard metals springs, the agariler claims.
Design a better chip
Aguler hoped that the new content would be the answer to the problem that stricken him, but even silicon springs did not strengthen the lidar system, as they needed to face the elements that came before them.
To make the lidar even stronger, Omnitron’s team targeted to design a more powerful MEMS chip by increasing the amount of force to control the device to control the mirror in the lidar array. They say they claim to achieve this – their chip can impose 10 times more force per unit area on an actuator that deploys a micromeir or other sensor component compared to the current industry standard, they say. This additional force allows for extremely valuable control in fine adjustment.
To reach this achievement, he had to excavate deeply.
Micromires of Omnitron can steal the lidar beam and used in data centers.Omnipotent
In this MEMS device, the mirror and its actuator are placed in the same silicone wafer. At the end of its non-mirror, the actuator is covered with small, closely spaced plates that fit between trenches in the wafer, such as two comb interlocking teeth. To move the mirror, the voltage is applied, and the electrostatic forces have angled the mirror in a specific position by moving the plates up and down the trenches. As the electric field draws across the trench sidewall,
The force that can be used to move the mirror is limited by the ratio of depth to the width of the trenches, which is called aspect ratio. Simply put, the deeper the trenches, the more electrostatic force can be applied to an actuator, which leads to a higher class of speed for the sensor. But creating deep, narrow trenches is a difficult effort. It was very important to control this limited factor.
Aguilar says that Omnitron was able to improve approximately 20: 1 aspect ratio, which he notes is specific to MEMS (other experts say that 30: 1 or 40: 1 is close to average these days), using 100: 1 through the use and prototypes of small universities across the United States. “This is really our main success. “It was through blood, sweat, tears and despair that we started this company.”
Startups in intention letters from automotive partners have achieved more than $ 800 million, says Aguler, and to prove that the 18 -month plan to prove is two months that it can produce its chips at the full demand rate.
Even after verifying production capabilities, technology will have to face “very difficult” security tests in thousands of hours of realistic conditions, such as vibration, thermal cycle and rain, before coming to the market, says Lee.
Savings power
Meanwhile, Omnitron is implementing its technique to solve a distinct problem faced by a separate industry. By 2030, AI data centers are expected to be required About 945 terravat hours To work – Japan consumes more today than the country of Japan. The problem is that “the way the data moves,” says the agariler. When the data is sent from one part of the data center to another, the optical signal is converted into electrical signals, resumed, and then returned to the optical signal to be sent on its way. This process, which is in a system called network switch, burns a large amount of electricity.
The solution of google is called ApolloData packets are to be placed as optical signals for the duration of your journey, which saves 40 percent of electricity, Company claimsDoes this using an array of mirrors to direct apollo data. The aguilar plans to make the process even more efficient using dense arrays of more powerful mirrors of omnitron. By doing this, each network can quadruple the amount of switch data, increasing the number of channels in each switch from 126 to 441.
Omnitron is still in a hurry to its data center implementation, so it is not yet clear to what extent this technique can actually improve on Google’s Apollo. However, after a “critical design review” in mid -September, one of the world’s top AI Hypersscalers has requested our mirrors for their next generation switch, “says Aguiler. “This is proof that Omnitron solves a problem that even the largest AI infrastructure companies cannot address the house.”
And there may be more applications to come. Aguilar says that Omnitron has felt from groups interested in detecting defense industry, space companies and methane. “It’s great that people knock at our door for this because I was just focusing on the lidar,” they say.
From your site articles
Related articles around web
