Issue link: https://iconnect007.uberflip.com/i/857644
74 The PCB Magazine • August 2017 So why is this important? First, don't look at HDI as something "just for mobile phones." When you hear someone capitulate to that way of reasoning, that person is not looking at HDI from a strategic point of view. And this is not to say that all fabricators should simply rush into HDI for to be fashionable! This is all about tech- nology and go-to-market strategy for the busi- ness. So think it through and assess one's capa- bilities. Don't expect that the fabricator already possesses the competencies to successfully im- plement an HDI manufacturing technology. That would be a mistake. Thus, it is important to remember that there are several disciplines to master including material selection, blind via and microvia formation, circuit formation, lam- ination and metallization. Even though micro- vias are the primary feature for HDI construc- tions, thinner dielectrics, ultra-thin copper foils and finer lines and spaces are key features as well. There is also a movement to coreless lam dielectrics and dielectric build-up materials in place of copper clad laminate. These last two are also enablers for finer lines and spaces. Basi- cally, getting into HDI requires changing one's thinking about PWB fabrication. And that must first start with getting familiar with standards. HDI Standards As one decides to adopt HDI, the IPC-2315 HDI Design Standard will be required reading. This critical document outlines the simplest ar- chitectures for HDI (Figure 2). Keep in mind that there are tradeoffs when designing interconnects with blinds vias. In ad- dition, PWBs with only through-holes designed in limit the density one can achieve. Without microvias, layer counts and costs will increase to support higher pin counts. PCB References 1. Holden, H., "Introduction to HDI Sub- strates and Microvia Technology," The Board Authority, Vol.2 No.1, April 2000, pp. 8–12. 2. IPC 2315. Michael Carano is VP of technology and business development for RBP Chemical Technology. To reach Carano, or read past columns, click here. MOVING INTO MICROVIAS: THE INTERACTION OF MATERIALS AND PROCESSES, PART 1 The best-performing artificial-intelligence sys- tems—in areas such as autonomous driving, speech recognition, computer vision, and automatic trans- lation—have come courtesy of software systems known as neural networks. But neural networks take up a lot of memory and consume a lot of power, so they usually run on servers in the cloud, which re- ceive data from desktop or mobile devices and then send back their analyses. Last year, MIT associate professor of electrical engineering and computer science Vivienne Sze and colleagues un- veiled a new, energy-efficient com- puter chip optimized for neural net- works, which could enable powerful artificial-intelligence systems to run locally on mobile devices. Now, Sze and her colleagues have approached the same problem from the opposite di- rection, with techniques for designing more energy- efficient neural networks. First, they developed an an- alytic method that can dete mine exactly how much power a neural network will consume. Then they used the method to evaluate new techniques for par- ing down neural networks so that they'll run more ef- ficiently on handheld devices. The researchers describe the work in a paper next week at the Computer Vision and Pattern Recognition Conference. In the pa- per, they report that the methods of- fered as much as a 73% reduction in power consumption over the stan- dard implementation of neural net- works, and as much as a 43% reduc- tion over the best previous method for paring the networks down. Bringing Neural Networks to Cellphones