DSSC Tape Out World First Machine Learning HDD Channel Chip
The two-dimensional magnetic recording (TDMR) is a technology that utilizes cross-track displaced multiple readers on the same data track in hard disk drives. The technology is designed to eliminate the inter-track interference as the data track pitch kept decreasing. In practice, however, this has been very challenging due to the drastic variation of the reader overlap as the slider moves across the disk surface. DSSC Ph.D. Student Yuwei Qin and Professor Jimmy Zhu had proposed a machine learning-based data detection channel to combat these challenges. Working in collaboration with Western Digital, a DSSC sponsoring company, they showed that the designed machine learning channel shows excellent performance with real data from real TDMR hard disk drives. Now, using TSMC 28nm technology, the researchers have designed and successfully taped out an ASCI chip, implementing their convolution neural network (CNN) based machine learning channel into real hardware. The taped-out chip with inputs from real hard disk drive read back signals demonstrated a performance that is superior to today’s hardware channel in state-of-the-art disk drives.