New Paradigm in Testing Heads & Media for HDD Dr. Lutz Henckels September 2010 1
WOW an amazing industry 40%+ per year aerial density growth Source: Coughlin Associates 2010 2
WOW an amazing industry Aerial Density Growth Driven by Innovations Source: Coughlin Associates 2010 3
WOW an amazing industry Price per disk units has dropped by factor of ~3 15% per year increase in disk units Source: Coughlin Associates 2010 4
WOW an amazing industry Highly Consolidated Industry Acquisition Targets? Source: Coughlin Associates 2010 Highly Undervalued Industry Enterprise Value/Revenue (ttm) Trailing P/E (ttm, intraday) EV/HPQ PE/HPQ WDC 0.37 4.40 48% 41% STX 0.46 3.54 60% 33% HIT 0.43 23.65 56% 219% DHR 2.22 20.56 HPQ 0.77 10.78 5
Two Approaches For HGA & Media Test Analog Measurements Digital Measurements -WOW 6
Analog Measurements Using Peak Detector to measure Average Amplitude Amplitude too high 75% 25% Rise Time too long 7
Analog Measurements Pros: Real time amplitude measurements Well established technology Cons: Measurements shortcomings: Not accurate in presence of noise Slow Spectral Measurements Limitations: No waveform capture, need scope to see signal No complex measurements such as Spectral SNR, Media noise, jitter, or 3D pulse profile Upgradability: Measurements fixed in hardware, not flexible enough for new test implementation and new technologies 8
Digital Measurements Precise Averaging To Extract True Signal 10 GSPS Acquisition and Average Profile 9
Digital Measurements Accurate Results in Presence of Noise True Amplitude 75% 25% True Rise Time High noise immunity, 1% measurement accuracy down to 12dB SNR 10
Why Digital To o see, measure, analyze, and store waveform 1. Accuracy: Higher accuracy due to removal of noise from signal 2. Flexibility: New Technologies & Test algorithms implemented by programming CPU, GPU, or FPGA 3. Make Complex Measurements: Average Pulse Profile Measurements Fast Spectrum Measurements Spectral SNR tests Eye Diagram and Jitter Analysis Media Defect Analysis 3D Pulse Profile for Nano-scale Magnetic Field Imaging 4. Display and Store Waveform: : Similar to Scope capability Remove the Mystery!! 11
Scopes can make all these digital measurements Great Engineering Tool 4 channels, High Bandwidth & Sample rate Scopes are too slow no Real-Time Parametric Measurements LeCroy 760 zi Agilent t Digital Signal Analyzer (DSA) Real Time Scope DSA 91304A 13GHz, 40Gs, 1Gpt $125,711 Production requires Throughput: Computing Power Must get measurement results instantaneously! 12
Building Digital Signal Analyzer for HDD Production High performance digitizer using DSA chipset DSA from major Oscilloscope vendor Up to 40 GSamples per second sampling rate 8-bit A/D Converter Up to 13 GHz Bandwidth 13
Digital Signal Analyzer for HDD Production High performance digitizer using Scope chipset Up to 40 GSamples per second sampling rate 8-bit A/D Converter Up to 13 GHz Bandwidth Combined with Computational Power Computational power for real time parametric test Four Altera Stratix IV FPGA s for equalization & Hardware-based signal processing 7.2 GSample/s parametric processing 14
Digital Signal Analyzer for HDD Production High performance digitizer using Scope chipset Up to 40 GSamples per second sampling rate 8-bit A/D Converter Up to 13 GHz Bandwidth Combined with Computational Power Computational power for real time parametric test Four Altera Stratix IV FPGA s s for Hardware-based signal processing 7.2GSample/s parametric processing Fast Data transfer PCI-Express to PC 8x + 16x = up to 9.6GByte/s data transfer 15
Digital Signal Analyzer for HDD Production High performance digitizer using Scope chipset Up to 40 GSamples per second sampling rate 8-bit A/D Converter Up to 13 GHz Bandwidth Combined with Computational Power Computational power for real time parametric test Fast Data transfer Four Altera Stratix IV FPGA s s for Hardware-based signal processing 7.2GSample/s parametric processing PCI-Express to PC 8x + 16x = up to 9.6GByte/s data transfer External parallel computers with up to 960 cores (4 in Scope) Dual Xeon six-core processor (Intel) = 12 cores at 3.33GHz Up to 4 Graphics Processors (GPU by NVIDIA or ATI) - 240 cores each 16
High performance digitizer using Scope chipset Up to 40 GSamples per second sampling rate 8-bit A/D Converter Up to 13 GHz Bandwidth Combined with Computational Power Computational power for real time parametric test Fast Data transfer Four Altera Stratix IV FPGA s s for Hardware-based signal processing 7.2GSample/s parametric processing PCI-Express to PC 8x + 16x = up to 9.6GByte/s data transfer External parallel computers with up to 960 cores (4 in Scope) Ultra-deep Memory Dual Xeon six-core processor (Intel) with 12 cores at 3.33GHz 4 Graphics Processors (GPU) - NVIDIA or ATI - with 240 cores each POWER: Ultra-long long Waveform Memory up to 32 Gigapoints/channel POWER: DDR3 memory for simultaneous data acquisition & processing Digital Signal Analyzer for HDD Production 17
Digital Signal Analyzer for HDD Production High performance digitizer using Scope chipset Up to 40 GSamples per second sampling rate 8-bit A/D Converter Up to 13 GHz Bandwidth Combined with Computational Power Computational power for real time parametric test Fast Data transfer Four Altera Stratix IV FPGA s s for Hardware-based signal processing 7.2GSample/s parametric processing PCI-Express to PC 8x + 16x = up to 9.6GByte/s data transfer External parallel computers with up to 960 cores (4 in Scope) Ultra-deep Memory Dual Xeon six-core processor (Intel) with 12 cores at 3.33GHz 4 Graphics Processors (GPU) - NVIDIA or ATI - with 240 cores each Ultra-long long Waveform Memory up to 32 Gigapoints on single channel DDR3 memory technology for simultaneous data acquisition and processing POWER: Double performance using two digitizers interleaved (80GSPS, 64GPoints, 32GHz) 18
Giga DSA for Production Digital Signal Analyzer (DSA) Giga DSA Sample Rate: 40 Gs/sec s/sec Bandwidth: 13 Ghz hz Memory: 32 Gpoints points Parametric Test: 7.2 Gs/s s/s Transfer Rate: 9.6 GB/sB/s PCI Express to PC 8x + 16x = 9.6GByte/s Multi module Extension 36GByte/s Four Processing FPGAs DDR3 Memory 32GBytes Power Supply 170W consumption ADC 13GHz, 36GSPS, 8 bit Input Amplifiers, 13 / 8 / 6.5 / 4GHz 19
Combine with Processing Power GPU CPU FPGA Up to 960 parallel cores on Four GPU cards Programming in CUDA C C & OpenCL Extensive Math libraries Add-on cards, easy upgrades Up to 12 parallel cores on Two Intel Xeon CPUs Programming in C/C++, Matlab, other high-level languages Easy upgrades Four Altera Stratix IV FPGAs 3600 multipliers Guzik implements processing units in Verilog 20
Combine with Software: Parametric Tests Real Time Parametric Tests Track Average Amplitude (TAA) Rise & Fall Time, Pulse Width, Roll-off off Asymmetry Tests: Pulse Width, Amplitude, Saturation Stability Tests: Amplitude, Sector Amplitude, Pulse Width Real time = Results are available immediately after last sample in last sector of revolution is acquired 21
FFT Analysis with GPU Spectral Signal-To To-Noise Ratio Measurements (Production Test) 22
Software: FFT Analysis with GPU 1.2 Gbytes/s @ 100KHz Resolution Bandwidth RBW Signal of 10 GSamples/sec and 4 GHz bandwidth 1 revolution at 7,200 RPM or 8.3 milliseconds/revolution FFT with 32,000 spectral lines takes 70 milliseconds 1.2 GB/s divided by 10 GSamples/sec equals 1/8 of real time FFT computed for 1 revolution in 70 milliseconds (8( 8 x 8.3ms) 4GHz bandwidth signal at.0001ghz RBW means 32,000 Spectral lines computed in 70 milliseconds Compared with Spectrum Analyzers 3000 times faster Spectrum Analyzer takes 32,000 x 8.3milliseconds or ~4 minutes 23
Sophisticated tests using CPU and GPU s 3D Pulse Profile for Nano-Scale Magnetic Field Imaging 100 ns / 1100 nm 600 nm Transition Curvature 10MHz pattern was written and then read and processed by 3D Pulse e Profile test. Red color indicates positive transitions, blue color indicates negative n transitions. 24
Sophisticated tests using CPU and GPU s 3D Pulse Profile for Nano-Scale Magnetic Field Imaging Track Offset (Cross-Track Direction) Head Flight (Along-the-Track Direction) Low-frequency perpendicular transitions, displayed as three-dimensional plot 25
Sophisticated tests using CPU and GPU s 3D Pulse Profile for Nano-Scale Magnetic Field Imaging Media Noise Test 26
Sophisticated tests using CPU and GPU s 3D Pulse Profile for Nano-Scale Magnetic Field Imaging Media Noise Test Eye Diagram and Jitter Analysis 27
Sophisticated tests using CPU and GPU s 3D Pulse Profile for Nano-Scale Magnetic Field Imaging Media Noise Test Eye Diagram and Jitter Analysis Media Defect Scanning 28
Sophisticated tests using CPU and GPU s 3D Pulse Profile for Nano-Scale Magnetic Field Imaging Media Noise Test Eye Diagram and Jitter Analysis Media Defect Scanning Adjacent Track Interference (ATI) Test Opening door for new Technologies Shingled Write Current Perpendicular to Plane (CPP) Heads TMR/GMR HAMR BPM DTR 29
Combine with Software Developer s s Kit User can create own test and measurement algorithms Can choose between processors: CPU (any language including MatLab) GPU ( CUDA( CUDA C, OpenGL, extensive processing libraries) FPGA (Guzik provides computation cores) 30
GSA is available as stand-alone alone Product GSA 20 Gsps GSA 20Gsps Scope 40GSPS / 13GHz Digitizer combined with Data Processor and 9.6GByte/s PCI-Express Link to PC Most of the measurements in Real Time 31
.or integrated to Guzik RWA / Spinstand System WITE Software, High TPI Spinstand, all integrated GSA Transparent Digital Measurement Solution Digital WITE plus more 32
for Head & Media Production Taking Mystery out of Test Results Capture, Measure, Analyze actual waveform Lower cost of ownership Price equivalent to scope ($50k to $100k) No need for Analog Hardware (HW) savings! no more upgrades Higher Throughput and Improved Yields Analysis of captured data - while moving Heads to Media 20 to 100 s s times faster analysis of complex algorithms Major HDD Vendor: 1% improvement in yields = $5M/yr savings Major HDD: 2 seconds reduction in test time = $20M/yr savings Enables New Technologies Smaller bits -11 MTPI, higher data rates -6Gb, more noise, more jitter More complex algorithms in production test Failure analysis with real data resolve issues much faster 33
34