Breaking AI's memory wall: former TSMC VP charts a path with HBM, optical links and 3D stacking
AI processor performance continues to improve, but memory bandwidth is growing more slowly, making data supply a major constraint on AI chip performance and energy efficiency. Efforts to shorten the distance between memory and compute units and increase die-to-die interconnect density are driving advances in HBM, advanced packaging and 3D integration.
Read original article ↗
Related Articles
Commentary: Samsung's US 2nm ramp faces no room for error
Samsung Electronics' 2nm mass-production capability at its Taylor, Texas fab has become critical to its efforts to close
TSMC to build another 3nm fab in Tainan
TSMC plans to build a new 3nm fab in Tainan, with production slated to start in the first half of 2028. The company is a
Microsoft's early adoption of TSMC's 2nm process, a US$2 billion manufacturing partnership between TSMC and GlobalFoundr
The road not taken Part 2 of 2: TSMC's first CTO said no to Intel twice but still wonders what if
Chenming Hu co-invented the FinFET transistor with colleagues at the University of California, Berkeley, in the 1990s an