And on the technology front does Cypress need to develop artificial intelligence and non-volatile memory?
"I don't follow the next shiny thing," said El-Khoury. "But I do see the significance of 'edge intelligence'," he added. "We need to be adding capabilities at the edge that can run on batteries. As to memory; we are the King of Memories. We are number one in automotive flash."
There are a number of companies coming through with variants of resistive RAM, magnetic RAM, phase-change memory, to provide non-volatility at geometries that flash is unable to achieve although often in the form of intellectual property as embedded memories, rather than discrete.
"The thing is we don't dabble; we focus. We can choose whether to develop something organically or license it in. It depends on how fast we want to do it. Machine learning could be an edge device in your smart home. But the use cases are lagging the technology." So, that suggests there is time to develop something organically? El-Khoury agrees but adds "It's about return on investment and the right time to market."
"MRAM is probably the most viable non-volatile memory but does not serve all use cases. It only goes to 85 degrees C and maybe 105 degrees C in a couple of years." These low operating temperature limits make it unsuitable for many automotive and industrial applications. "Cypress has embedded flash memory. We have SONOS [silicon-oxide-nitride-oxide-silicon] and eCT [embedded charge trap] memory. At 40nm we offer 8Mbytes of embedded flash. At more advanced nodes customers tend to use external flash, El-Khoury said.
There is an argument that customers use external discrete memories because no non-volatile memory has yet emerged with the right balance of speed, density and endurance and applicable at 20nm and below, but that when it does it will sweep the field.
Next: I won't bet the ranch