Title: A Reconfigurable FPGABased Readback Signal Generator for HardDrive Read Channel Simulator
1A Reconfigurable FPGABased Readback Signal
Generator for HardDrive Read Channel Simulator
- Jinghuan Chen Jaekyun Moon Kia Bazargan
- ECE Department
- University of Minnesota
2Outline
- Problem domain hard disk readback signal
simulation - Modeling HDD read
- FPGA implementation
- Hardware platform
- Design components
- Random number generator
- Results and conclusion
Outline
3Outline
- Problem domain hard disk readback signal
simulation - Modeling HDD read
- FPGA implementation
- Hardware platform
- Design components
- Random number generator
- Results and conclusion
Outline
4Reconfiguration Benefits
- Reuse hardware
- Saves area
- Leaves room for data recovery circuit
- Intermediate data?
- Partial runtime reconfiguration (future work)
- No need to synthesize each configuration(synthesi
ze individual noise components,assemble at
runtime) - Can cover all possible noise combinations
- No need to transfer intermediate data to on-board
mem
5Hard Disk Technology
- Disk metal or glass coated with magnetic thin
film - Write induce magnetic directions
- Read
- pick up magnetic flux
- Pre-amplification, analog/digital filtering, data
recovery
V
Readback pulses
Read Process
Write Process
6Transition Noise
- Trend pack more bits ? more noise
- Transition noise (data dependent)
- Imperfect boundary between bit cells
- ? pulse shifts in time domain
- ? pulse shape deforms
Transition position jitter
Pulse width variation
7Read / Write Nonlinear Factors
- Head magneto-resistant
- Non-linear in nature different /- amplitude
- Partial erasure
- Irregularity of disk spinning speed
- Head not 100 on the track? cross-talk from
neighboring track - Electronics noise
8Outline
- Problem domain hard disk readback signal
simulation - Modeling HDD read
- FPGA implementation
- Hardware platform
- Design components
- Random number generator
- Results and conclusion
Outline
9Simulating HDD Read / Write Why?
- High demand for reliability
- Very low bit error rate (BER) ? long simulation
- Desktop 10 out of 1013.
- Server 10 out of 1015.
- High throughput
- Push data recovery techniques to the limits
- Need to understand effects of
- Advanced, novel signal processing techniques
- Encoding schemes(e.g., turbo codes, low density
parity check) - Hard disk noise diff from other comm systems
- Make dedicated signal generator for simulation
- Previous work software simulation takes DAYS
10Benefits of Modeling in FPGA
- Speed
- Instead of days, hours
- Hardware integration into data recovery model
- Reconfiguration
- Ability to add/remove different noise components
- Ability to tune parameters of a particular noise
factor - Runtime reconfiguration
- Done by time-multiplexing different
configurationson FPGA - Pre-synthesized modules
- Possible partial runtime reconfiguration
- Example neighboring track interference might
bevary among different tracks
11Modeling Noise Factors
- Inter-symbol interference (neighboring bits)
12Modeling Noise Factors (cont.)
- Head non-linearity F(x)
- Amplitude loss due to partial erasure
- Disk spinning irregularities
- Modeled as a slow-paced random-walk
- Inter-track interference
- Weighted sum of two uncorrelated signal
generators(each representing a track) - Electronics noise
- Additive white bandlimited gaussian noise (AWGN)
13Outline
- Problem domain hard disk readback signal
simulation - Modeling HDD read
- FPGA implementation
- Hardware platform
- Design components
- Random number generator
- Results and conclusion
Outline
14FPGA Hardware Platform
- FPGA board
- Annapolis Micro Systems FireBird board
- Xilinx XCV1000E FPGA
- 18MB of RAM
- PCI bus interface
MEM
MEM
MEM
XCV1000E
Host Computer
PCI Controller
MEM
MEM
15Outline
- Problem domain hard disk readback signal
simulation - Modeling HDD read
- FPGA implementation
- Hardware platform
- Design components
- Random number generator
- Results and conclusion
Outline
16Design Hierarchy and Behavior
Board
FPGA
PCI Bus
HOST CPU
On-board RAM
17Design Behavior
- Restructuring
- FPGA configuration (add / remove noise
components) - Model different hard disks
- Initialization / Re-initialization
- Lookup table initializations (change parameters)
- Pulse lookup tables
- Gaussian conversion tables
- Multiplier lookup tables
18Signal Generator Components
Input
- Data / interfering track
- Data main data track
- Interfering neighboring track
- Both pipelined
- Independent random number generators
- Weighted sum to model interference
19Data Track Components
Data Track
LHCAPRNG
NLTS
WERR
20Outline
- Problem domain hard disk readback signal
simulation - Modeling HDD read
- FPGA implementation
- Hardware platform
- Design components
- Random number generator
- Results and conclusion
Outline
21Uniform Random Number Generation
- Classic method LFSR (Linear Feedback Shift Reg)
- Large area
- Feedback loop makes routing hard
- Linear hybrid cellular automata (LHCA)
- Smaller in area ( factor of 5 reduction)
- Needs a number of initialization cycles
- Null boundary ? no feedback loop
- Not every bit position random? use one bit per n
registers - Implementation use Rule 90 and Rule 150 to
combine neighboring bits
22LHCA Implementation
1
i-1
i
N
I1
0
0
- For each bit position, carefully chooseRule 90
or Rule 150
23Example Modeling WERR
- WERR caused by irregular disk spinning speed
- Modeled as a slow-paced random walk
- Add one write error in every NWERR clock cycles
(1,-1)
RNG
1
ò
0
0
CLK k.NWERR(k1,2,3, ...)
24Non-uniform Random Number
- Use a uniform RNG, lookup the values
ofnon-uniform random numbers from a table - Lookup table
- User-defined (Gaussian, etc.)
- Cumulative distribution function (CDM)
- Encoder
- Round-off to L bits (quantization error depends
on L)
25Example Gaussian PDF
26Outline
- Problem domain hard disk readback signal
simulation - Modeling HDD read
- FPGA implementation
- Hardware platform
- Design components
- Random number generator
- Results and conclusion
Outline
27Sample Output
1.0
0.5
0.0
-0.5
-1.0
0
200
400
600
800
1000
28Resource Utilization and Speedup
- Configuration 0 uses 99 of chip area
- Clock speed 70 MHz
- Speedup compared to software 4 hr vs. 50,000 hr
29Floorplan
30Floorplan
31Reconfiguration Benefits
- Reuse hardware
- Saves area
- Leaves room for data recovery circuit
- Intermediate data?
- Partial runtime reconfiguration (future work)
- No need to synthesize each configuration(synthesi
ze individual noise components,assemble at
runtime) - Can cover all possible noise combinations
- No need to transfer intermediate data to on-board
mem - Implementation
- Constrain the floorplan
- Use JBits
32Constrained Floorplan for Partial Reconfiguration
33Placement for Partial Reconfig.