Everyone knows that Nvidia (NVDA) currently has the best AI chips. However, there is another segment of the semiconductor industry that is just as important to AI as Nvidia’s GPUs (graphics processing units).
The demand for advanced packaging is growing rapidly because the computers training AI programs need high-bandwidth memory (HBM) interfaces. There is no point in having GPUs to do the advanced computing if you can’t transfer data to and from them quickly. And the faster the data can be transferred, the faster the models can be trained.
For higher-bandwidth memory, designers have been favoring what is referred to as a stacked design, in which several memory chips are stacked on top of one another and then packaged with a GPU. All these chips need to be connected by several tiny wires, a process known as advanced packaging.
The big problem with advanced packaging is that it creates an extra layer of complexity. The need to connect more chips adds more potential malfunction points—and that is where the technology from this under-the-radar AI play comes in.
Let me fill you in on all the details.
Camtek
The company is Camtek (CAMT), and its metrology and 3D sensing systems are crucial. Its sensors—coupled with the company’s proprietary software—detect faults and stop manufacturers from sending out any faulty products.
Camtek has developed sensors that shine light onto the semiconductors and then analyze what bounces back, a bit like sonar. By using its proprietary software, its systems can build up a 3D picture of the chip, identify its faults, and ultimately improve the yield of the manufacturer.
The main chiplet and high-bandwidth manufacturers are Taiwan Semiconductor (TSM), Intel (INTC), and the two Korean powerhouses, Samsung and SK Hynix. While Camtek doesn’t reveal the exact deals it signs, it says that 70% of its business comes from…
Continue reading at INVESTORSALLEY.com
